{"models":[{"name":"timm/aimv2_1b_patch14_224.apple_pt","title":"AIMv2 1B/14 (224)","source":"timm/aimv2_1b_patch14_224.apple_pt","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Faimv2_1b_patch14_224.apple_pt&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/aimv2_1b_patch14_224.apple_pt","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":37.7006,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":42.208,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":26.5,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":17.55,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":17.03,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"f0e84c28e009480e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/aimv2_1b_patch14_336.apple_pt","title":"AIMv2 1B/14 (336)","source":"timm/aimv2_1b_patch14_336.apple_pt","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Faimv2_1b_patch14_336.apple_pt&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/aimv2_1b_patch14_336.apple_pt","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":80.9993,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":83.5975,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":12.3,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":17.55,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":17.03,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"27a39f86891b7c9b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/aimv2_1b_patch14_448.apple_pt","title":"AIMv2 1B/14 (448)","source":"timm/aimv2_1b_patch14_448.apple_pt","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Faimv2_1b_patch14_448.apple_pt&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/aimv2_1b_patch14_448.apple_pt","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":146.756,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":150.295,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":6.8,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":17.57,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":17.05,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"3d6fbb62fee6b5c6","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/aimv2_3b_patch14_224.apple_pt","title":"AIMv2 3B/14 (224)","source":"timm/aimv2_3b_patch14_224.apple_pt","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Faimv2_3b_patch14_224.apple_pt&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/aimv2_3b_patch14_224.apple_pt","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":89.3087,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":92.3516,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":11.2,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":38.11,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":37.59,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"bbd27a47febf419f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/aimv2_3b_patch14_336.apple_pt","title":"AIMv2 3B/14 (336)","source":"timm/aimv2_3b_patch14_336.apple_pt","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Faimv2_3b_patch14_336.apple_pt&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/aimv2_3b_patch14_336.apple_pt","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":171.811,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":178.311,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":5.8,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":38.13,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":37.61,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"ad55dba91fd2095d","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/aimv2_3b_patch14_448.apple_pt","title":"AIMv2 3B/14 (448)","source":"timm/aimv2_3b_patch14_448.apple_pt","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Faimv2_3b_patch14_448.apple_pt&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/aimv2_3b_patch14_448.apple_pt","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":310.278,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":321.438,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3.2,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":38.13,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":37.61,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"6715748cb9295d05","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/aimv2_huge_patch14_224.apple_pt","title":"AIMv2 Huge/14 (224)","source":"timm/aimv2_huge_patch14_224.apple_pt","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Faimv2_huge_patch14_224.apple_pt&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/aimv2_huge_patch14_224.apple_pt","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":23.5287,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":24.9102,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":42.5,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":9.91,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":9.38,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"491a170283e6b474","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/aimv2_huge_patch14_336.apple_pt","title":"AIMv2 Huge/14 (336)","source":"timm/aimv2_huge_patch14_336.apple_pt","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Faimv2_huge_patch14_336.apple_pt&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/aimv2_huge_patch14_336.apple_pt","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":47.4907,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":50.0944,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":21.1,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":9.91,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":9.39,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"e7c84dee95e87302","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/aimv2_huge_patch14_448.apple_pt","title":"AIMv2 Huge/14 (448)","source":"timm/aimv2_huge_patch14_448.apple_pt","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Faimv2_huge_patch14_448.apple_pt&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/aimv2_huge_patch14_448.apple_pt","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":84.0714,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":86.9355,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":11.9,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":9.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":9.4,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"d3f7209ad2b4a414","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/aimv2_large_patch14_224.apple_pt","title":"AIMv2 Large/14 (224)","source":"timm/aimv2_large_patch14_224.apple_pt","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Faimv2_large_patch14_224.apple_pt&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/aimv2_large_patch14_224.apple_pt","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.58204,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":10.2346,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":104.3,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4.79,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":4.26,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"9c69f51e3f482431","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/aimv2_large_patch14_336.apple_pt","title":"AIMv2 Large/14 (336)","source":"timm/aimv2_large_patch14_336.apple_pt","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Faimv2_large_patch14_336.apple_pt&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/aimv2_large_patch14_336.apple_pt","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":22.7265,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":24.2504,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":44,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4.79,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":4.27,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"f591de23c3739cb7","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/aimv2_large_patch14_448.apple_pt","title":"AIMv2 Large/14 (448)","source":"timm/aimv2_large_patch14_448.apple_pt","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Faimv2_large_patch14_448.apple_pt&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/aimv2_large_patch14_448.apple_pt","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":41.6904,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":42.5996,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":24,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4.8,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":4.28,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"13bca9aaf049aadb","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"nvidia/alpamayo2-edge","upstream":"nvidia/Alpamayo2-Edge","title":"Alpamayo 2 Edge","source":"nvidia/Alpamayo2-Edge","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=nvidia%2FAlpamayo2-Edge&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"nvidia/alpamayo2-edge","hardware":"thor","precision":"NVFP4 / FP8","metrics":[{"title":"Latency","metrics":[{"description":"Per decision, median","value":91.9,"unit":"ms","higherIsBetter":false,"digits":1,"baseline":603.7,"baselineLabel":"NVIDIA PyTorch BF16","headline":true},{"description":"Per decision, p90","value":140.3,"unit":"ms","higherIsBetter":false,"digits":1,"baseline":840.6,"baselineLabel":"NVIDIA PyTorch BF16"},{"description":"Decisions per second","value":10.9,"unit":"Hz","higherIsBetter":true,"digits":1,"baseline":1.7,"baselineLabel":"NVIDIA PyTorch BF16"}]},{"title":"Accuracy","subtitle":"PhysicalAI-AV held-out validation · 378 clips","metrics":[{"description":"minADE6 at 5 s","value":0.464,"unit":"m","higherIsBetter":false,"digits":3,"baseline":0.508,"baselineLabel":"NVIDIA PyTorch BF16","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per decision","value":12.45,"unit":"J","higherIsBetter":false,"digits":2,"baseline":51.87,"baselineLabel":"NVIDIA PyTorch BF16","improvementWord":"less","headline":true},{"description":"Power while deciding, mean","value":117.6,"unit":"W","higherIsBetter":false,"digits":1,"baseline":74.5,"baselineLabel":"NVIDIA PyTorch BF16"},{"description":"Power while deciding, peak","value":150.2,"unit":"W","higherIsBetter":false,"digits":1,"baseline":137.9,"baselineLabel":"NVIDIA PyTorch BF16"},{"description":"Power at idle","value":30,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":7.31,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":15.94,"baselineLabel":"NVIDIA PyTorch BF16"}]},{"title":"Size","metrics":[{"description":"Container size, on disk","value":1.07,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":11.35,"baselineLabel":"NVIDIA PyTorch BF16","headline":true},{"description":"Model size","value":2.37,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":4.95,"baselineLabel":"NVIDIA PyTorch BF16"}]}],"labels":["Image","Text","VLA"],"build":"202611102057c0ba","updatedAt":"2026-11-10T23:25:38.000Z"},{"model":"nvidia/alpamayo2-edge","hardware":"thor","variant":"huggingface_fp16_tensorrt-edgellm","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Per decision, median","value":489.3,"unit":"ms","higherIsBetter":false,"digits":1,"headline":true},{"description":"Per decision, p90","value":713.2,"unit":"ms","higherIsBetter":false,"digits":1},{"description":"Decisions per second","value":2,"unit":"Hz","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"PhysicalAI-AV held-out validation · 189 clips","metrics":[{"description":"minADE6 at 5 s","value":0.436,"unit":"m","higherIsBetter":false,"digits":3,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per decision","value":62.02,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Power while deciding, mean","value":103.4,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while deciding, peak","value":163,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":30,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":10.61,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLA"],"runnable":false,"updatedAt":"2026-11-10T23:25:38.000Z"},{"model":"nvidia/alpamayo2-edge","hardware":"thor","variant":"huggingface_nvfp4_embedl_tensorrt-edgellm","precision":"NVFP4 / FP8","metrics":[{"title":"Latency","metrics":[{"description":"Per decision, median","value":91.9,"unit":"ms","higherIsBetter":false,"digits":1,"baseline":603.7,"baselineLabel":"NVIDIA PyTorch BF16","headline":true},{"description":"Per decision, p90","value":140.3,"unit":"ms","higherIsBetter":false,"digits":1,"baseline":840.6,"baselineLabel":"NVIDIA PyTorch BF16"},{"description":"Decisions per second","value":10.9,"unit":"Hz","higherIsBetter":true,"digits":1,"baseline":1.7,"baselineLabel":"NVIDIA PyTorch BF16"}]},{"title":"Accuracy","subtitle":"PhysicalAI-AV held-out validation · 378 clips","metrics":[{"description":"minADE6 at 5 s","value":0.464,"unit":"m","higherIsBetter":false,"digits":3,"baseline":0.508,"baselineLabel":"NVIDIA PyTorch BF16","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per decision","value":12.45,"unit":"J","higherIsBetter":false,"digits":2,"baseline":51.87,"baselineLabel":"NVIDIA PyTorch BF16","improvementWord":"less","headline":true},{"description":"Power while deciding, mean","value":117.6,"unit":"W","higherIsBetter":false,"digits":1,"baseline":74.5,"baselineLabel":"NVIDIA PyTorch BF16"},{"description":"Power while deciding, peak","value":150.2,"unit":"W","higherIsBetter":false,"digits":1,"baseline":137.9,"baselineLabel":"NVIDIA PyTorch BF16"},{"description":"Power at idle","value":30,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":7.31,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":15.94,"baselineLabel":"NVIDIA PyTorch BF16"}]},{"title":"Size","metrics":[{"description":"Container size, on disk","value":1.07,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":11.35,"baselineLabel":"NVIDIA PyTorch BF16","headline":true},{"description":"Model size","value":2.37,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":4.95,"baselineLabel":"NVIDIA PyTorch BF16"}]}],"labels":["Image","Text","VLA"],"build":"202611102057c0ba","updatedAt":"2026-11-10T23:25:38.000Z"},{"model":"nvidia/alpamayo2-edge","hardware":"h200","precision":"BF16","metrics":[{"title":"Latency","subtitle":"per decision through the server · 10 requests, median · synthetic 576x320 frames at NVIDIA's schedule, sent as raw RGB pixels as a car's cameras hand them over","metrics":[{"description":"Decision","value":116.5,"unit":"ms","higherIsBetter":false,"headline":true},{"description":"Model","value":111.6,"unit":"ms","higherIsBetter":false},{"description":"Preprocessing","value":4.8,"unit":"ms","higherIsBetter":false},{"description":"Request handling","value":0.1,"unit":"ms","higherIsBetter":false}]},{"title":"Memory","subtitle":"the server's GPU memory (nvidia-smi, every 50 ms) at its highest during 45 live decisions on real drive frames; the model is its checkpoint's files","metrics":[{"description":"Peak GPU memory","value":7.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Model size","value":4.95,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLA"],"build":"9b96b8fb88e3cbc3","updatedAt":"2026-11-10T23:25:38.000Z"}]},{"name":"timm/bat_resnext26ts.ch_in1k","title":"BAT ResNeXt-26-TS","source":"timm/bat_resnext26ts.ch_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fbat_resnext26ts.ch_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/bat_resnext26ts.ch_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.777289,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.860168,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1267,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.05,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.03,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d2cbb28fed9e09e3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/bat_resnext26ts.ch_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.06804,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.08185,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":480.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.05,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.03,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"22b4646ff83f77c7","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/bat_resnext26ts.ch_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.50099,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.51074,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":284.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.05,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.03,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2f3c6a5c6f22af16","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/bat_resnext26ts.ch_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.974438,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.980804,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1024.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.05,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.03,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.56,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":38.6,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"202610201336c292","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/bat_resnext26ts.ch_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.3127,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.31952,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":761.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.05,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.03,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ed19eebf4ae0ea84","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/beit_base_patch16_224.in22k_ft_in22k_in1k","title":"BEiT Base/16 (224)","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fbeit_base_patch16_224.in22k_ft_in22k_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/beit_base_patch16_224.in22k_ft_in22k_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.457,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.461,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":685.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":1.07,"unit":"%","headline":true},{"description":"Top-5","value":3.56,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/beit_large_patch16_224.in22k_ft_in22k_in1k","title":"BEiT Large/16 (224)","source":"timm/beit_large_patch16_224.in22k_ft_in22k_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fbeit_large_patch16_224.in22k_ft_in22k_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/beit_large_patch16_224.in22k_ft_in22k_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.55377,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.6394,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":219.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":85.06,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.52,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.18,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.65,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"fb50d263ddf5893c","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/beit_large_patch16_384.in22k_ft_in22k_in1k","title":"BEiT Large/16 (384)","source":"timm/beit_large_patch16_384.in22k_ft_in22k_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fbeit_large_patch16_384.in22k_ft_in22k_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/beit_large_patch16_384.in22k_ft_in22k_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":21.4642,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":21.9224,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":46.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":86.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.19,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c383d2542ffd110c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/beit_large_patch16_384.in22k_ft_in22k_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":25.9873,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":26.0688,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":38.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":86.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.19,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8cd412b2b4365282","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/beit_large_patch16_384.in22k_ft_in22k_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":61.0864,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":61.552,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":16.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":86.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.19,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5c41a36cd1de3d0f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/beit_large_patch16_384.in22k_ft_in22k_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":20.2239,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":21.165,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":49.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":86.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.19,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.27,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.75,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"a549961772b5ec6e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/beit_large_patch16_384.in22k_ft_in22k_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":23.4873,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":23.6753,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":42.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":86.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.19,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ebb58c03329332d3","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/beit_large_patch16_512.in22k_ft_in22k_in1k","title":"BEiT Large/16 (512)","source":"timm/beit_large_patch16_512.in22k_ft_in22k_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fbeit_large_patch16_512.in22k_ft_in22k_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/beit_large_patch16_512.in22k_ft_in22k_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":46.932,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":47.0532,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":21.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":87.3,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.35,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"151d46383d691c09","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/beit_large_patch16_512.in22k_ft_in22k_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":58.3701,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":58.4136,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":17,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":87.3,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.35,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8c34366c478f60e4","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/beit_large_patch16_512.in22k_ft_in22k_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":142.495,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":142.625,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":87.3,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.35,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c8b3bdfa69642b4d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/beit_large_patch16_512.in22k_ft_in22k_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":57.0882,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":59.6135,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":17.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":87.3,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.35,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.37,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.85,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"512654c67b8a23b9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/beit_large_patch16_512.in22k_ft_in22k_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":45.6146,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":46.0757,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":21.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":87.3,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.35,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"027410040cb14001","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/beit3_base_patch16_224.indomain_pt","title":"BEiT-3 Base/16 (224)","source":"timm/beit3_base_patch16_224.indomain_pt","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fbeit3_base_patch16_224.indomain_pt&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/beit3_base_patch16_224.indomain_pt","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.38797,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.39978,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":719.6,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.38,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.86,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"aa0c4d9d06c45a5a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/beit3_large_patch16_224.in22k_ft_in1k","title":"BEiT-3 Large/16 (224, IN-22k → IN-1k)","source":"timm/beit3_large_patch16_224.in22k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fbeit3_large_patch16_224.in22k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/beit3_large_patch16_224.in22k_ft_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.88558,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.65832,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":2.97595,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.75525,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":342.5,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":270.544,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.93,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.99,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"641dc45c4b1a7e54","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/beit3_large_patch16_224.in22k_ft_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.77074,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":9.8064,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":101.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.93,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.99,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1c6300177837fcc9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/beit3_large_patch16_224.in22k_ft_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":21.0904,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":21.1343,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":47.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.93,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.99,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7a9600fc6afb9be0","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/beit3_large_patch16_224.in22k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.93893,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.17667,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":4.00488,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.41302,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":253.8,"unit":"qps","digits":1,"baseline":193.112,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":85.93,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.99,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.56,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.04,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"d5ce97771431304d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/beit3_large_patch16_224.in22k_ft_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.97969,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.03503,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":251.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.93,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.99,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4129b8d6b8273623","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/beit3_large_patch16_224.indomain_pt","title":"BEiT-3 Large/16 (224, indomain, pt)","source":"timm/beit3_large_patch16_224.indomain_pt","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fbeit3_large_patch16_224.indomain_pt&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/beit3_large_patch16_224.indomain_pt","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.97624,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.19093,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":4.05341,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.36859,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":251.4,"unit":"qps","digits":1,"baseline":192.576,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.55,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.03,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"2a20b2157f08b454","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/botnet26t_256.c1_in1k","title":"BoTNet-26-T (256)","source":"timm/botnet26t_256.c1_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fbotnet26t_256.c1_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/botnet26t_256.c1_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.350142,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.413513,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2770.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.89,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.38,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"74d94077d75723d3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/botnet26t_256.c1_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.95701,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.973267,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1038.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.89,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.38,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"49a3e328d2d62ac4","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/botnet26t_256.c1_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.78417,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.80493,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":557.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.89,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.38,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"68d1ab4427b23bd8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/botnet26t_256.c1_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.374045,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.375214,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2662.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.89,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.38,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.69,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.17,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"45f74ce6798f9d4a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/botnet26t_256.c1_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.588212,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.591614,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1696.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.89,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.38,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"592b88cb3c6e99de","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/caformer_s36.sail_in1k","title":"CAFormer-S36","source":"timm/caformer_s36.sail_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcaformer_s36.sail_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/caformer_s36.sail_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.72445,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.72778,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":579.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.73,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.59,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.97,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.44,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"f2877d2f40336af4","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/cait_m36_384.fb_dist_in1k","title":"CaiT-M36 (384)","source":"timm/cait_m36_384.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcait_m36_384.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/cait_m36_384.fb_dist_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":23.9544,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":24.1914,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":41.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.89,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.27,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"de65d925aa6eb4d3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cait_m36_384.fb_dist_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":55.8134,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":56.0327,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":17.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.89,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.27,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"297756a1bb55ccf8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cait_m36_384.fb_dist_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":142.358,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":142.43,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.89,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.27,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"90ee21e32188da3b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cait_m36_384.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":25.0157,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":26.0471,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":40,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.89,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.27,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.56,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.04,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"a8135008c067fc17","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cait_m36_384.fb_dist_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":27.2896,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":27.4552,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":36.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.89,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.27,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a99c21bbad0416ef","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/cait_m48_448.fb_dist_in1k","title":"CaiT-M48 (448)","source":"timm/cait_m48_448.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcait_m48_448.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/cait_m48_448.fb_dist_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":56.2339,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":56.5181,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":17.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.45,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.42,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9d9dafc84407fdfe","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cait_m48_448.fb_dist_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":129.748,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":130.22,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":7.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.45,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.42,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8c37f9fd647f6500","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cait_m48_448.fb_dist_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":335.476,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":335.883,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.45,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.42,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f698fb45b71e9bc8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cait_m48_448.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":86.169,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":86.9458,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":11.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":85.45,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.42,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4.52,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":4,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"a8cb996fe0b1a618","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cait_m48_448.fb_dist_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":62.6588,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":62.9717,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":16,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.45,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.42,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d438496f3327f887","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/cait_s24_224.fb_dist_in1k","title":"CaiT-S24 (224)","source":"timm/cait_s24_224.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcait_s24_224.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/cait_s24_224.fb_dist_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.07698,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.17981,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":322.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.05,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"bef7099909d976de","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cait_s24_224.fb_dist_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.41549,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.42749,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":155.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.05,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"97ad27325fa28790","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cait_s24_224.fb_dist_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":13.7034,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":13.7208,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":72.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.05,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d967e3a7ff698503","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cait_s24_224.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.22711,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.29382,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":448.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.05,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.05,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.53,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"3b07d5bf1c42d68d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cait_s24_224.fb_dist_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.85238,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.87561,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":350.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.05,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d9bacab5dfefea90","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/cait_s24_384.fb_dist_in1k","title":"CaiT-S24 (384)","source":"timm/cait_s24_384.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcait_s24_384.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/cait_s24_384.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":10.1182,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":10.343,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":98.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.18,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.16,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.05,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.53,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"5009ce52e3b98cc9","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/cait_s36_384.fb_dist_in1k","title":"CaiT-S36 (384)","source":"timm/cait_s36_384.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcait_s36_384.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/cait_s36_384.fb_dist_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":16.8091,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":16.9034,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":59.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.91,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8a853a4bfad6e496","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cait_s36_384.fb_dist_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":41.9882,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":42.2092,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":23.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.91,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"960fbc9dbf16c4c8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cait_s36_384.fb_dist_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":102.794,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":103.346,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":9.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.91,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"cc9e46cd7c654165","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cait_s36_384.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":14.7566,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":15.1921,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":67.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.91,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.29,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.77,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"0793724f6d913924","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cait_s36_384.fb_dist_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":21.1405,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":21.2236,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":47.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.91,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9f9c426c166e168a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/cait_xs24_384.fb_dist_in1k","title":"CaiT-XS24 (384)","source":"timm/cait_xs24_384.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcait_xs24_384.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/cait_xs24_384.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.83079,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":9.07727,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":113.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.8,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.18,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.82,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.3,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"894b66051b5a6741","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/cait_xxs24_224.fb_dist_in1k","title":"CaiT-XXS24 (224)","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcait_xxs24_224.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/cait_xxs24_224.fb_dist_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.476,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.478,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":676.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":75.68,"unit":"%","headline":true},{"description":"Top-5","value":93.03,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cait_xxs24_224.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.47054,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.47217,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":679.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":75.64,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.09,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.66,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.14,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"891261ffd7f2685f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/cait_xxs36_224.fb_dist_in1k","title":"CaiT-XXS36 (224)","source":"timm/cait_xxs36_224.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcait_xxs36_224.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/cait_xxs36_224.fb_dist_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.75629,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.86176,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":263.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.48,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.24,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ed9992d1811319a9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cait_xxs36_224.fb_dist_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.97264,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.99744,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":142.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.48,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.24,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c3850adb29bd0cfa","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cait_xxs36_224.fb_dist_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":14.4004,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":14.5364,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":69.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.48,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.24,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"75a63a97eb2bb213","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cait_xxs36_224.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.1645,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.16602,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":461.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.48,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.24,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.72,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.2,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"0e2b5a2c1401cdbd","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cait_xxs36_224.fb_dist_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.90627,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.92596,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":344,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.48,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.24,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"606aea138038b346","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/cait_xxs36_384.fb_dist_in1k","title":"CaiT-XXS36 (384)","source":"timm/cait_xxs36_384.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcait_xxs36_384.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/cait_xxs36_384.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.53403,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":8.69751,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":117.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.87,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.08,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.72,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.2,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"c9882c654a765b97","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/coat_mini.in1k","title":"CoaT Mini","source":"timm/coat_mini.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcoat_mini.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/coat_mini.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.7844,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.0437,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":357.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.2,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.39,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f0cd06d6c2a3db31","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_mini.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.26319,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":8.36829,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":120.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.2,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.39,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9bd6e6674ee7d306","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_mini.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":15.6821,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":15.7747,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":63.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.2,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.39,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"237859bacfb46651","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_mini.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.63061,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.73047,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":379.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.2,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.39,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.68,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.16,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"3ab71483d89c9f5a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_mini.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.03801,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.0553,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":247.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.2,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.39,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"312afec6266380b1","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/coat_small.in1k","title":"CoaT Small","source":"timm/coat_small.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcoat_small.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/coat_small.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.4517,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.59558,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":288.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.34,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.17,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b138bc0c40191995","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_small.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.55788,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":9.65137,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":104.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.34,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.17,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"fdacd81dba170726","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_small.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":18.7178,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":18.8237,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":53.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.34,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.17,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7b1aad12a8fa952e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_small.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.50692,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.60651,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":285,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.34,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.17,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.85,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.32,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"a6096a295eb0854e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_small.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.0031,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.02869,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":199.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.34,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.17,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"744bd91aa22dbfd4","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/coat_tiny.in1k","title":"CoaT Tiny","source":"timm/coat_tiny.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcoat_tiny.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/coat_tiny.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.65461,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.84506,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":374.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.31,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9dcbf2095da65590","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_tiny.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.19116,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":7.23169,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":138.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.31,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1ce62bc1c8307400","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_tiny.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":13.1227,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":13.2568,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":75.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.31,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"28e17e23c2da666c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_tiny.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.16406,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.16937,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":461.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.31,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.6,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":85.2,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"686877c82e007122","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_tiny.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.40508,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.42224,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":293.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.31,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"78c749e8b63252ae","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/coat_lite_medium.in1k","title":"CoaT-Lite Medium","source":"timm/coat_lite_medium.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcoat_lite_medium.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/coat_lite_medium.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.86394,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.01251,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":347.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.54,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.67,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2a8e49e7c8b44f21","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_lite_medium.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.84867,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.91156,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":145.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.54,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.67,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"234de6cc2f228a60","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_lite_medium.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":12.8055,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":12.9247,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":77.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.54,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.67,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7b886e4b18de0c10","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_lite_medium.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.61292,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.67902,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":382.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.54,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.67,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.04,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"491dcdfa34fc8231","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_lite_medium.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.55399,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.57373,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":281.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.54,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.67,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e72a202020bf1df2","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/coat_lite_medium_384.in1k","title":"CoaT-Lite Medium (384)","source":"timm/coat_lite_medium_384.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcoat_lite_medium_384.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/coat_lite_medium_384.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.50863,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.64764,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":179.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.33,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"491243d5f672381e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_lite_medium_384.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":13.3645,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":13.4496,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":74.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.33,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8bf26cad394b25fd","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_lite_medium_384.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":28.19,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":28.4146,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":35.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.33,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d3d0c6275f7ad065","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_lite_medium_384.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.48183,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.59875,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":182.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.33,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.04,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"ede69250760a9460","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_lite_medium_384.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.19404,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":7.22839,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":139,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.33,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"46fc93190201cd7c","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/coat_lite_mini.in1k","title":"CoaT-Lite Mini","source":"timm/coat_lite_mini.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcoat_lite_mini.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/coat_lite_mini.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.818648,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.921875,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1196.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.24,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1e1bea428e6fd50c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_lite_mini.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.96089,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.97314,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":507.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.24,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"51e1177e15b61413","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_lite_mini.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.65127,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.67352,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":272.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.24,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"af105c575a984a1f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_lite_mini.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.721415,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.723083,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1382.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.24,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.65,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.13,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"6ba753fb559971ac","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_lite_mini.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.06864,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.07111,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":934.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.24,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"de2840fdbdd229d4","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/coat_lite_small.in1k","title":"CoaT-Lite Small","source":"timm/coat_lite_small.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcoat_lite_small.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/coat_lite_small.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.50927,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.69478,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1.61896,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.8103,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":655.4,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":581.904,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.57,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"712d60dd0c9abd53","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_lite_small.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.48943,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.49872,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":285,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.57,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2fee69b275a4fca3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_lite_small.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.66409,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.68921,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":149.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.57,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"179542ce4bf10f33","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_lite_small.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.29413,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.2973,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":771.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.57,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.75,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.23,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"a9464235ad795122","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_lite_small.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.88082,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.88391,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":531.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.57,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"03d58c965a8d0831","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/coat_lite_tiny.in1k","title":"CoaT-Lite Tiny","source":"timm/coat_lite_tiny.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcoat_lite_tiny.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/coat_lite_tiny.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.794917,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.894348,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1231,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.31,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.74,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7cb05e93923cdc32","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_lite_tiny.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.10846,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.12061,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":471.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.31,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.74,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"fe55ed01e1227449","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_lite_tiny.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.42077,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.45142,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":290.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.31,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.74,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d98aa66e1a857905","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_lite_tiny.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.676181,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.67804,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1475.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.31,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.74,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":70.1,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"7d1326577ddf7590","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/coat_lite_tiny.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.975589,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.978943,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1024,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.31,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.74,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9acc9e31589adfbd","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/coatnet_3_rw_224.sw_in12k","title":"CoAtNet-3 (224, rw)","source":"timm/coatnet_3_rw_224.sw_in12k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcoatnet_3_rw_224.sw_in12k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/coatnet_3_rw_224.sw_in12k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.87043,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.95233,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":258.3,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.36,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.84,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"a7303311735f99a0","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convit_tiny.fb_in1k","title":"ConViT Tiny","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvit_tiny.fb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convit_tiny.fb_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":499.268,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":502.625,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":1996.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":22.08,"unit":"%","headline":true},{"description":"Top-5","value":43.43,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnext_atto.d2_in1k","title":"ConvNeXt Atto","source":"timm/convnext_atto.d2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnext_atto.d2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnext_atto.d2_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.377945,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.451019,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2585.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.17,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.66,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"df59e941a6dd1a74","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_atto.d2_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.24294,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.24774,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":799.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.17,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.66,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"48625c873bae6cab","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_atto.d2_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.06589,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.08636,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":481.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.17,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.66,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"cc4ac7a7b4e01a31","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_atto.d2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.399335,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.400146,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2493.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":75.17,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.66,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.57,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":44.7,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"0b18772fde9615ea","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_atto.d2_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.634979,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.636658,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1572.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.17,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.66,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9e3f4a4dc9e2fcaa","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnext_atto_ols.a2_in1k","title":"ConvNeXt Atto (ols)","source":"timm/convnext_atto_ols.a2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnext_atto_ols.a2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnext_atto_ols.a2_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.37556,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.446747,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2613.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.58,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5688dfeffc601950","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_atto_ols.a2_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.26144,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.26605,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":787.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.58,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"36b12fb116cbb3c4","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_atto_ols.a2_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.10648,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.11566,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":472,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.58,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"18859ff7c5709f8c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_atto_ols.a2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.400724,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.402344,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2485.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":75.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.58,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.57,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":44.8,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"1602ac5a895fbc80","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_atto_ols.a2_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.643552,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.645386,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1551.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.58,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0e800715d33dfe66","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/convnext_base","title":"ConvNeXt Base","source":"torchvision/convnext_base","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=facebook%2Fconvnext-base-224&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/convnext_base","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":7.50150921754539,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":133.30650819718855,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":82.588,"unit":"%","digits":2,"baseline":84.062,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":96.236,"unit":"%","digits":2,"baseline":96.87,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/convnext_base.clip_laion2b_augreg_ft_in1k","title":"ConvNeXt Base (CLIP, AugReg, IN-1k fine-tuned)","source":"timm/convnext_base.clip_laion2b_augreg_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnext_base.clip_laion2b_augreg_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnext_base.clip_laion2b_augreg_ft_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.82694,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.04311,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1.9408,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.15997,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":540.4,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":483.773,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.78,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.62,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"843716d444ca1377","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_base.clip_laion2b_augreg_ft_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.99237,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.00629,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":249.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.78,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.62,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"29614f5463af5ac6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_base.clip_laion2b_augreg_ft_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.40915,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":8.43652,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":118.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.78,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.62,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"fdb0b6a61a21ac20","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_base.clip_laion2b_augreg_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.86588,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.16188,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.91949,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.23657,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":535.5,"unit":"qps","digits":1,"baseline":462.256,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":85.78,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.62,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.53,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"6a2ddfc803de3d86","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_base.clip_laion2b_augreg_ft_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.25961,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.27045,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":442.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.78,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.62,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a780d46a16cf8a93","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnext_base.clip_laion2b","title":"ConvNeXt Base (CLIP)","source":"timm/convnext_base.clip_laion2b","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnext_base.clip_laion2b&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnext_base.clip_laion2b","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.9037,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.16924,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.91113,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.22607,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":524.9,"unit":"qps","digits":1,"baseline":460.682,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.52,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"fd240a0e7a9fb97f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnext_femto.d1_in1k","title":"ConvNeXt Femto","source":"timm/convnext_femto.d1_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnext_femto.d1_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnext_femto.d1_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.407453,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.485901,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2387.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.07,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.5,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f7268843c066ddd6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_femto.d1_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.33305,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.33801,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":745.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.07,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.5,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9cf692977c00cfc2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_femto.d1_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.20383,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.22693,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":451.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.07,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.5,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6fdab62b5457589f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_femto.d1_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.429087,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.430298,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2322.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.07,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.5,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.58,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":62.6,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"1703f3cf98467457","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_femto.d1_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.689347,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.691711,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1448.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.07,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.5,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"bd7eab4a20f10403","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnext_femto_ols.d1_in1k","title":"ConvNeXt Femto (ols)","source":"timm/convnext_femto_ols.d1_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnext_femto_ols.d1_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnext_femto_ols.d1_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.468636,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.469543,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2126.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.79,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.31,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.58,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":62.8,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"c00ec11ada0ee1f6","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/convnext_large","title":"ConvNeXt Large","source":"torchvision/convnext_large","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=facebook%2Fconvnext-large-224&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/convnext_large","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":8.416139479959384,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":118.81932356054845,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":83.062,"unit":"%","digits":2,"baseline":84.414,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":96.372,"unit":"%","digits":2,"baseline":96.976,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/convnext_large.dinov3_lvd1689m","title":"ConvNeXt Large (DINOv3)","source":"timm/convnext_large.dinov3_lvd1689m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnext_large.dinov3_lvd1689m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnext_large.dinov3_lvd1689m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.9709,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.3661,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":3.04858,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.42395,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":336.4,"unit":"qps","digits":1,"baseline":296.948,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.74,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.22,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"240c450f7a435d09","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnext_large_mlp.clip_laion2b_augreg_ft_in1k","title":"ConvNeXt Large (mlp, CLIP, IN-1k fine-tuned)","source":"timm/convnext_large_mlp.clip_laion2b_augreg_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnext_large_mlp.clip_laion2b_augreg_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnext_large_mlp.clip_laion2b_augreg_ft_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.67844,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.23848,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":2.78033,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.33246,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":369.9,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":305.207,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":87.07,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.17,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2c554174641881f8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_large_mlp.clip_laion2b_augreg_ft_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.41002,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":7.3563,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":6.4245,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":7.37354,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":155.2,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":135.235,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":87.07,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.17,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"235ce8268f131327","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_large_mlp.clip_laion2b_augreg_ft_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":13.6602,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":13.6831,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":72.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":87.07,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.17,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1c5500b0aceab37c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_large_mlp.clip_laion2b_augreg_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.19923,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.28125,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":312.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":87.07,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.17,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.79,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.27,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"73822bc728e96d08","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_large_mlp.clip_laion2b_augreg_ft_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.54357,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.57385,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":282.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":87.07,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.17,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4e2b52637e922077","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnext_large_mlp.clip_laion2b_augreg","title":"ConvNeXt Large (mlp, CLIP)","source":"timm/convnext_large_mlp.clip_laion2b_augreg","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnext_large_mlp.clip_laion2b_augreg&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnext_large_mlp.clip_laion2b_augreg","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.1602,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.23175,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":316.3,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.78,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.26,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"fafbb7f2f843c21a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnext_large.fb_in1k","title":"ConvNeXt Large (timm)","source":"timm/convnext_large.fb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnext_large.fb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnext_large.fb_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.58486,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.10864,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":2.67908,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.21741,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":382,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":318.059,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.89,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.72,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"06975c7fe3820325","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_large.fb_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.86253,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":6.77868,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":5.8927,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":6.79919,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":169.7,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":146.755,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.89,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.72,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"712a266b705da199","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_large.fb_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":12.5715,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":12.5964,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":79.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.89,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.72,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d6389d1c4e63b03c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_large.fb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.89436,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.30052,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":2.96698,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.36566,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":345.3,"unit":"qps","digits":1,"baseline":302.85,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.89,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.72,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.76,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.24,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"49c7065c68b23030","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_large.fb_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.40817,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.41663,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":293.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.89,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.72,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"385cbcaa97dadb1f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnext_nano.d1h_in1k","title":"ConvNeXt Nano","source":"timm/convnext_nano.d1h_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnext_nano.d1h_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnext_nano.d1h_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.538606,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.655091,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.614563,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.720764,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1806.9,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1474.91,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.31,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.1,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f0f4dfb6c7e0b1c0","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_nano.d1h_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.87465,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.8808,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":530.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.31,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.1,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4edc4b3022f87986","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_nano.d1h_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.47113,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.47986,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":286.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.31,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.1,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"56d2d0c55aee4219","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_nano.d1h_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.649112,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.65155,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1537.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.31,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.1,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.7,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.18,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"f523dad2187c8295","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_nano.d1h_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.01089,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.01337,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":988.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.31,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.1,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"912725b8f3c7f640","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnext_nano_ols.d1h_in1k","title":"ConvNeXt Nano (ols)","source":"timm/convnext_nano_ols.d1h_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnext_nano_ols.d1h_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnext_nano_ols.d1h_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.547317,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.687568,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.621155,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.754639,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1776.3,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1411.06,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.44,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.07,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8aa7119161c52d4e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_nano_ols.d1h_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.92626,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.93243,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":516.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.44,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.07,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0b18f9348e16d881","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_nano_ols.d1h_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.52912,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.54163,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":281.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.44,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.07,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7fe9b8a205651445","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_nano_ols.d1h_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.633932,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.636353,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1573.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.44,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.07,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.7,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.18,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"1a81deaac77dad44","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_nano_ols.d1h_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.01486,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.01764,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":984.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.44,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.07,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b6b22dc905b8d49c","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnext_pico.d1_in1k","title":"ConvNeXt Pico","source":"timm/convnext_pico.d1_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnext_pico.d1_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnext_pico.d1_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.409692,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.479187,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2378,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.17,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.38,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"29b07df780e1f350","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_pico.d1_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.48185,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.48865,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":670.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.17,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.38,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"116fbf6e79073211","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_pico.d1_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.51132,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.52032,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":396,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.17,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.38,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0d6666d0dc59a2af","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_pico.d1_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.487608,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.488708,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2045.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.17,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.38,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.63,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.11,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"9dd97ea4f3885eaa","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_pico.d1_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.785288,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.78775,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1271.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.17,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.38,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d1e303f295fb2d03","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnext_pico_ols.d1_in1k","title":"ConvNeXt Pico (ols)","source":"timm/convnext_pico_ols.d1_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnext_pico_ols.d1_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnext_pico_ols.d1_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.405173,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.485779,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2400.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.97,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.43,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4de1fb158a10a6c3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_pico_ols.d1_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.50765,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.51422,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":659.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.97,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.43,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f93a249d577350d4","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_pico_ols.d1_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.5654,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.57355,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":387.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.97,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.43,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"70cbf8243701d322","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_pico_ols.d1_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.487335,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.488708,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2045.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.97,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.43,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.63,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.11,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"ea0eb41a602f7ae7","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_pico_ols.d1_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.796462,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.798492,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1253.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.97,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.43,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0dd47cedf951b897","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/convnext_small","title":"ConvNeXt Small","source":"torchvision/convnext_small","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=facebook%2Fconvnext-small-224&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/convnext_small","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":7.30948974378407,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":136.8084551798423,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":82.232,"unit":"%","digits":2,"baseline":83.616,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":95.918,"unit":"%","digits":2,"baseline":96.65,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/convnext_small.dinov3_lvd1689m","title":"ConvNeXt Small (DINOv3)","source":"timm/convnext_small.dinov3_lvd1689m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnext_small.dinov3_lvd1689m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnext_small.dinov3_lvd1689m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.4638,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.46704,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":682.5,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.09,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.56,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"4ecb412fabef4b51","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnext_small.fb_in1k","title":"ConvNeXt Small (timm)","source":"timm/convnext_small.fb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnext_small.fb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnext_small.fb_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.52286,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.63458,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":645.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.93,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.4,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3258c235dd3ea50d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_small.fb_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.19085,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.20605,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":311.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.93,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.4,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"acda6ac8b3d1fe52","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_small.fb_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.2509,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.27136,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":159.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.93,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.4,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"82cf1cae6d8aa0b0","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_small.fb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.4997,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.50287,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":666.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.93,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.4,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.09,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.57,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"52c0f9967a61817e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_small.fb_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.80828,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.81641,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":552.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.93,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.4,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7166dc26720bc127","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/convnext_tiny","title":"ConvNeXt Tiny","source":"torchvision/convnext_tiny","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fconvnext_tiny&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/convnext_tiny","hardware":"thor","precision":"FP8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.951172,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.22559,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Throughput","value":1044.67,"unit":"qps","digits":1,"baseline":809.618,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNette validation · 3,925 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":88.07643312101911,"unit":"%","digits":2,"baseline":88.2547770700637,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Top-5","value":99.36305732484077,"unit":"%","digits":2,"baseline":99.36305732484077,"baselineLabel":"Baseline TensorRT"}]},{"title":"Memory","metrics":[{"description":"Peak device memory","value":31,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":58,"baselineLabel":"Baseline TensorRT","improvementWord":"smaller","headline":true}]},{"title":"Size","metrics":[{"description":"Engine size","value":29.5,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":56,"baselineLabel":"Baseline TensorRT","improvementWord":"smaller","headline":true}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/convnext_tiny","hardware":"orin-nano-super","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":19.2415,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":51.7698,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNette validation · 3,925 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":84.73885350318471,"unit":"%","digits":2,"baseline":87.21019108280255,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":98.49681528662421,"unit":"%","digits":2,"baseline":99.3375796178344,"baselineLabel":"FP32 model"}]},{"title":"Size","metrics":[{"description":"Engine size","value":44,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"torchvision/convnext_tiny","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.665527,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.824219,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Throughput","value":1498.7,"unit":"qps","digits":1,"baseline":1214.01,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNette validation · 3,925 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":88.12738853503184,"unit":"%","digits":2,"baseline":88.2547770700637,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Top-5","value":99.38853503184714,"unit":"%","digits":2,"baseline":99.38853503184714,"baselineLabel":"Baseline TensorRT"}]},{"title":"Memory","metrics":[{"description":"Peak device memory","value":37,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":59,"baselineLabel":"Baseline TensorRT","improvementWord":"smaller","headline":true}]},{"title":"Size","metrics":[{"description":"Engine size","value":36.3,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":59.2,"baselineLabel":"Baseline TensorRT","improvementWord":"smaller","headline":true}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnext_tiny.dinov3_lvd1689m","title":"ConvNeXt Tiny (DINOv3)","source":"timm/convnext_tiny.dinov3_lvd1689m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnext_tiny.dinov3_lvd1689m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnext_tiny.dinov3_lvd1689m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.686514,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.789237,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.68869,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.790894,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1453.3,"unit":"qps","digits":1,"baseline":1264.74,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.84,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.32,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"e7557be947e41ec8","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnext_tiny_hnf.a2h_in1k","title":"ConvNeXt Tiny (hnf)","source":"timm/convnext_tiny_hnf.a2h_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnext_tiny_hnf.a2h_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnext_tiny_hnf.a2h_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.908025,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.0008,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1.0069,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.08777,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1075.1,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":974.843,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.81,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.64,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"42d3a4063a055ee5","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_tiny_hnf.a2h_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.5799,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.59393,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":385.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.81,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.64,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6d7e53b6981c40b2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_tiny_hnf.a2h_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.94456,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.95801,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":201.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.81,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.64,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e119ff8e84078f82","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_tiny_hnf.a2h_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.884042,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.886597,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1129.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.81,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.64,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.85,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.33,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e81690e7a28f9e93","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_tiny_hnf.a2h_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.38092,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.38373,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":723.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.81,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.64,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"43499000ce6d8651","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnext_tiny.fb_in1k","title":"ConvNeXt Tiny (timm)","source":"timm/convnext_tiny.fb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnext_tiny.fb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnext_tiny.fb_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.859801,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.960449,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1132,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.82,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.78,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9113ec4bbab54d9f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_tiny.fb_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.78738,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.7937,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":556.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.82,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.78,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8d26c50cbaa1a426","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_tiny.fb_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.53713,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.54761,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":281.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.82,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.78,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"64f262ee3e569231","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_tiny.fb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.67855,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.801275,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.680664,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.804565,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1470.8,"unit":"qps","digits":1,"baseline":1245.49,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.82,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.78,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.85,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.33,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"f272c7eafdb08b6b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_tiny.fb_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.04875,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.05518,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":952.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.82,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.78,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"48475569edbaa025","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnextv2_atto.fcmae_ft_in1k","title":"ConvNeXt V2 Atto (FCMAE, IN-1k fine-tuned)","source":"timm/convnextv2_atto.fcmae_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnextv2_atto.fcmae_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnextv2_atto.fcmae_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.612263,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.613586,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1629.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.19,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.88,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.57,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.05,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"430a50226d4d3a3a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnextv2_atto.fcmae","title":"ConvNeXt V2 Atto (FCMAE)","source":"timm/convnextv2_atto.fcmae","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnextv2_atto.fcmae&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnextv2_atto.fcmae","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.644907,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.646179,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1546.5,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.56,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":41.1,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image"],"build":"a143ceeea7c3499f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnextv2_base.fcmae","title":"ConvNeXt V2 Base (FCMAE)","source":"timm/convnextv2_base.fcmae","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnextv2_base.fcmae&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnextv2_base.fcmae","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.80543,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.862,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":356.2,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.52,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"a6de2f420bd21aaa","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnextv2_femto.fcmae_ft_in1k","title":"ConvNeXt V2 Femto (FCMAE, IN-1k fine-tuned)","source":"timm/convnextv2_femto.fcmae_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnextv2_femto.fcmae_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnextv2_femto.fcmae_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.666842,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.669067,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1496.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.91,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.67,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.07,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"3f44c85542c38c64","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnextv2_femto.fcmae","title":"ConvNeXt V2 Femto (FCMAE)","source":"timm/convnextv2_femto.fcmae","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnextv2_femto.fcmae&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnextv2_femto.fcmae","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.685283,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.686554,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1455.9,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.58,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":58.3,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image"],"build":"38ca3c4456cc6e04","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnextv2_huge.fcmae","title":"ConvNeXt V2 Huge (FCMAE)","source":"timm/convnextv2_huge.fcmae","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnextv2_huge.fcmae&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnextv2_huge.fcmae","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.38085,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":10.5889,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":9.62933,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":11.0253,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":106.6,"unit":"qps","digits":1,"baseline":94.4254,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":7.94,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":7.42,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"3fa487fcc229ff6f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnextv2_large.fcmae","title":"ConvNeXt V2 Large (FCMAE)","source":"timm/convnextv2_large.fcmae","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnextv2_large.fcmae&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnextv2_large.fcmae","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.33101,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.4137,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":230.8,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.74,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.22,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"ebdaf636a4b45e50","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnextv2_nano.fcmae","title":"ConvNeXt V2 Nano (FCMAE)","source":"timm/convnextv2_nano.fcmae","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnextv2_nano.fcmae&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnextv2_nano.fcmae","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.01831,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.02087,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":980.5,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.69,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.17,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"05c253c9c56496af","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnextv2_pico.fcmae_ft_in1k","title":"ConvNeXt V2 Pico (FCMAE, IN-1k fine-tuned)","source":"timm/convnextv2_pico.fcmae_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnextv2_pico.fcmae_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnextv2_pico.fcmae_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.794974,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.801208,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1255.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.61,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.79,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.64,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.12,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"00c78a9ea900e512","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnextv2_pico.fcmae","title":"ConvNeXt V2 Pico (FCMAE)","source":"timm/convnextv2_pico.fcmae","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnextv2_pico.fcmae&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnextv2_pico.fcmae","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.795105,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.799072,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1254.9,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.62,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":102,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image"],"build":"0f803c46bdfa3a81","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnextv2_tiny.fcmae","title":"ConvNeXt V2 Tiny (FCMAE)","source":"timm/convnextv2_tiny.fcmae","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnextv2_tiny.fcmae&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnextv2_tiny.fcmae","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.22373,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.23328,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":816.1,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.84,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.32,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"814fd57472087b5a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnext_xlarge.fb_in22k_ft_in1k","title":"ConvNeXt XLarge","source":"timm/convnext_xlarge.fb_in22k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnext_xlarge.fb_in22k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnext_xlarge.fb_in22k_ft_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.38619,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.43615,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":3.50342,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.6554,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":293.2,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":224.225,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":86.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.06,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"389729b7b06f5889","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_xlarge.fb_in22k_ft_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.13861,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":9.19058,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":8.1842,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":9.24536,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":122.2,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":108.249,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":86.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.06,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a826b0a3db1662ea","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_xlarge.fb_in22k_ft_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":17.3904,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":17.4437,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":57.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":86.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.06,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d2514bf8cd4f5d71","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_xlarge.fb_in22k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.54944,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.55042,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":4.67017,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.70227,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":219.7,"unit":"qps","digits":1,"baseline":180.119,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":86.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.06,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4.49,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.97,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"2c2fa2e9059e2e50","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_xlarge.fb_in22k_ft_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.28255,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.34192,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":233.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":86.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.06,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ec5fa1f609f6b3e5","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnext_xxlarge.clip_laion2b_rewind","title":"ConvNeXt XXLarge (CLIP, rewind)","source":"timm/convnext_xxlarge.clip_laion2b_rewind","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnext_xxlarge.clip_laion2b_rewind&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnext_xxlarge.clip_laion2b_rewind","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.89089,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":11.4505,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":9.10706,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":12.1898,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":112.5,"unit":"qps","digits":1,"baseline":87.319,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":10.06,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":9.54,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"5796e234c58ed5bf","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnext_xxlarge.clip_laion2b_soup_ft_in1k","title":"ConvNeXt XXLarge (CLIP, soup, IN-1k fine-tuned)","source":"timm/convnext_xxlarge.clip_laion2b_soup_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnext_xxlarge.clip_laion2b_soup_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnext_xxlarge.clip_laion2b_soup_ft_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.68317,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":9.84006,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":6.83044,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":10.1023,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":149.1,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":100.859,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":88.18,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.61,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a3ef5cd5f35bed8f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_xxlarge.clip_laion2b_soup_ft_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":15.9725,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":21.4799,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":15.9924,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":21.5547,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":62.3,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":46.3206,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":88.18,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.61,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1b7c0f7273957361","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_xxlarge.clip_laion2b_soup_ft_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":39.1382,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":39.1938,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":25.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":88.18,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.61,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8e815132526cc462","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_xxlarge.clip_laion2b_soup_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.77315,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":11.5106,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":8.99573,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":12.0697,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":114,"unit":"qps","digits":1,"baseline":86.8645,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":88.18,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.61,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":10.06,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":9.54,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"a2defb70b63f881d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_xxlarge.clip_laion2b_soup_ft_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.79431,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":9.85913,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":102.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":88.18,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.61,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"81f4defaf9bfca4d","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnext_zepto_rms_ols.ra4_e3600_r224_in1k","title":"ConvNeXt Zepto (rms, ols)","source":"timm/convnext_zepto_rms_ols.ra4_e3600_r224_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnext_zepto_rms_ols.ra4_e3600_r224_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnext_zepto_rms_ols.ra4_e3600_r224_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.381145,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.460907,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2565,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.56,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ece655afc77e98a1","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_zepto_rms_ols.ra4_e3600_r224_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.63228,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.63763,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":609,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.56,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9797c10f5fe92cd3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_zepto_rms_ols.ra4_e3600_r224_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.66011,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.6723,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":373.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.56,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c03ed07934243e65","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_zepto_rms_ols.ra4_e3600_r224_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.55072,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.552002,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1810.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":72.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.56,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.55,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":26.5,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"9e110271e5296205","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_zepto_rms_ols.ra4_e3600_r224_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.787167,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.788879,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1268.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.56,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9626858e07193530","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/convnext_zepto_rms.ra4_e3600_r224_in1k","title":"ConvNeXt Zepto (rms)","source":"timm/convnext_zepto_rms.ra4_e3600_r224_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvnext_zepto_rms.ra4_e3600_r224_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convnext_zepto_rms.ra4_e3600_r224_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.359169,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.434631,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2733,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":70.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.99,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"194561a076e4485b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_zepto_rms.ra4_e3600_r224_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.5669,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.57507,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":634.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":70.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.99,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a6d794eb144e44c3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_zepto_rms.ra4_e3600_r224_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.54803,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.55878,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":390.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":70.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.99,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a35b7b41bb449c32","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_zepto_rms.ra4_e3600_r224_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.549638,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.550415,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1814.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":70.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.99,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.55,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":26.3,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"4923f6801e3c4dcb","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/convnext_zepto_rms.ra4_e3600_r224_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.742548,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.745239,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1344.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":70.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.99,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8141fba7f7031094","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"facebook/convnext-tiny-224","title":"ConvNeXt-Tiny (HF)","source":"facebook/convnext-tiny-224","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=facebook%2Fconvnext-tiny-224&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"facebook/convnext-tiny-224","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.689941,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.824707,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Throughput","value":1446,"unit":"qps","digits":1,"baseline":1213,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNette validation · 3,925 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":88.38216560509554,"unit":"%","digits":2,"baseline":89.09554140127388,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Top-5","value":99.54140127388536,"unit":"%","digits":2,"baseline":99.59235668789809,"baselineLabel":"Baseline TensorRT"}]},{"title":"Memory","metrics":[{"description":"Peak device memory","value":40,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":59,"baselineLabel":"Baseline TensorRT","improvementWord":"smaller","headline":true}]},{"title":"Size","metrics":[{"description":"Engine size","value":37.4,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":59.1,"baselineLabel":"Baseline TensorRT","improvementWord":"smaller","headline":true}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"nvidia/Cosmos-Reason1-7B","title":"Cosmos-Reason1 7B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=nvidia%2FCosmos-Reason1-7B&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"nvidia/Cosmos-Reason1-7B","hardware":"thor","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":41.5,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":16.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":3516,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":6719,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":50.8,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":16.6,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":57.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":58.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":83.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":84.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":43.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":44.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":45.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":46.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.79,"unit":"J","higherIsBetter":false,"digits":2,"baseline":3.87,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":0.56,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":0.26,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":74.3,"unit":"W","higherIsBetter":false,"digits":1,"baseline":62.3,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":92.9,"unit":"W","higherIsBetter":false,"digits":1,"baseline":95.6,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":23.3,"unit":"W","higherIsBetter":false,"digits":1,"baseline":22.2,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":8.66,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":17.4,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":7.48,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":6.47,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"a4f9bd9eae5e71ec","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"nvidia/Cosmos-Reason1-7B","hardware":"thor","variant":"huggingface_fp16_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":16.1,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":6719,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":16.6,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":58.4,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"GSM8K","value":84.4,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"IFEval","value":44.4,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"MMLU-Pro","value":46.5,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":3.87,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.26,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":62.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":95.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":22.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":17.4,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"nvidia/Cosmos-Reason1-7B","hardware":"thor","variant":"huggingface_int4_embedl_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":41.5,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":16.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":3516,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":6719,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":50.8,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":16.6,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":57.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":58.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":83.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":84.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":43.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":44.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":45.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":46.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.79,"unit":"J","higherIsBetter":false,"digits":2,"baseline":3.87,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":0.56,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":0.26,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":74.3,"unit":"W","higherIsBetter":false,"digits":1,"baseline":62.3,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":92.9,"unit":"W","higherIsBetter":false,"digits":1,"baseline":95.6,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":23.3,"unit":"W","higherIsBetter":false,"digits":1,"baseline":22.2,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":8.66,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":17.4,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":7.48,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":6.47,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"a4f9bd9eae5e71ec","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"nvidia/Cosmos-Reason1-7B","hardware":"orin","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":27.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":2254,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":34.2,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":57.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":58.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":83.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":84.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":43.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":44.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":45.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":46.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.69,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.59,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":46.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":62.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":4.85,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":7.4,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":6.48,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"nvidia/Cosmos-Reason1-7B","hardware":"orin","variant":"huggingface_int4_embedl_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":27.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":2254,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":34.2,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":57.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":58.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":83.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":84.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":43.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":44.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":45.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":46.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.69,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.59,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":46.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":62.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":4.85,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":7.4,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":6.48,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"e49d7fed8e499dae","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"nvidia/Cosmos-Reason2-2B","title":"Cosmos-Reason2 2B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=nvidia%2FCosmos-Reason2-2B&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"nvidia/Cosmos-Reason2-2B","hardware":"thor","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":149.5,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":63.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":13875,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":32508,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":179.2,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":65.1,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":58.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":61.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":79,"unit":"%","higherIsBetter":true,"digits":1,"baseline":80.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":56,"unit":"%","higherIsBetter":true,"digits":1,"baseline":59.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":41.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":45.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.43,"unit":"J","higherIsBetter":false,"digits":2,"baseline":0.97,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":2.35,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":1.03,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":63.5,"unit":"W","higherIsBetter":false,"digits":1,"baseline":61.8,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":70.8,"unit":"W","higherIsBetter":false,"digits":1,"baseline":70.6,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":22.2,"unit":"W","higherIsBetter":false,"digits":1,"baseline":22.9,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":4.54,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":6.12,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":3.69,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":2.68,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"67c24d8f04e36185","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"nvidia/Cosmos-Reason2-2B","hardware":"thor","variant":"huggingface_fp16_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":63.5,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":32508,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":65.1,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":61.7,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"GSM8K","value":80.3,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"IFEval","value":59.5,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"MMLU-Pro","value":45.2,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.97,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.03,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":61.8,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":70.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":22.9,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":6.12,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"nvidia/Cosmos-Reason2-2B","hardware":"thor","variant":"huggingface_int4_embedl-flashhead_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":149.5,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":63.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":13875,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":32508,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":179.2,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":65.1,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":58.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":61.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":79,"unit":"%","higherIsBetter":true,"digits":1,"baseline":80.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":56,"unit":"%","higherIsBetter":true,"digits":1,"baseline":59.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":41.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":45.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.43,"unit":"J","higherIsBetter":false,"digits":2,"baseline":0.97,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":2.35,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":1.03,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":63.5,"unit":"W","higherIsBetter":false,"digits":1,"baseline":61.8,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":70.8,"unit":"W","higherIsBetter":false,"digits":1,"baseline":70.6,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":22.2,"unit":"W","higherIsBetter":false,"digits":1,"baseline":22.9,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":4.54,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":6.12,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":3.69,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":2.68,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"67c24d8f04e36185","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"nvidia/Cosmos-Reason2-2B","hardware":"orin","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":92.9,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":8428,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":111.9,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":59.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":61.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":78.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":80.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":56.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":59.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":42.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":45.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.42,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":2.36,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":39.4,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":47.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":0.95,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":3.6,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":2.68,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"e99219df0e084e92","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"nvidia/Cosmos-Reason2-2B","hardware":"orin","variant":"huggingface_int4_embedl-flashhead_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":92.9,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":8428,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":111.9,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":59.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":61.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":78.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":80.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":56.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":59.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":42.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":45.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.42,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":2.36,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":39.4,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":47.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":0.95,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":3.6,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":2.68,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"e99219df0e084e92","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"nvidia/Cosmos-Reason2-2B","hardware":"orin-nx-16gb","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":45.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":3242,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":58.6,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":59.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":61.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":79.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":80.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":57.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":59.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":41.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":45.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.42,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":2.36,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":19.4,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":22.8,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":8.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":2.67,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":3.6,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":2.68,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"nvidia/Cosmos-Reason2-2B","hardware":"orin-nx-16gb","variant":"huggingface_int4_embedl-flashhead_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":45.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":3242,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":58.6,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":59.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":61.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":79.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":80.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":57.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":59.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":41.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":45.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.42,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":2.36,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":19.4,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":22.8,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":8.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":2.67,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":3.6,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":2.68,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"nvidia/Cosmos-Reason2-2B","hardware":"orin-nano-super","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":47,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":2852,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":63.4,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":59.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":61.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":79.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":80.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":58,"unit":"%","higherIsBetter":true,"digits":1,"baseline":59.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":42.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":45.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.43,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":2.3,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":20.4,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":22,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":7.9,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":4.04,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":3.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":2.67,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"603643b847e8a010","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"nvidia/Cosmos-Reason2-2B","hardware":"orin-nano-super","variant":"huggingface_int4_embedl-flashhead_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":47,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":2852,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":63.4,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":59.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":61.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":79.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":80.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":58,"unit":"%","higherIsBetter":true,"digits":1,"baseline":59.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":42.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":45.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.43,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":2.3,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":20.4,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":22,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":7.9,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":4.04,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":3.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":2.67,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"603643b847e8a010","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"nvidia/Cosmos-Reason2-8B","title":"Cosmos-Reason2 8B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=nvidia%2FCosmos-Reason2-8B&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"nvidia/Cosmos-Reason2-8B","hardware":"thor","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":37.8,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":15.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":3244,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":6777,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":46,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":15.8,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":77.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":79.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":93.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":94.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":78.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":79.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":61.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":64.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.89,"unit":"J","higherIsBetter":false,"digits":2,"baseline":4.15,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":0.53,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":0.24,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":71.4,"unit":"W","higherIsBetter":false,"digits":1,"baseline":63.7,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":91.7,"unit":"W","higherIsBetter":false,"digits":1,"baseline":99.5,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":23,"unit":"W","higherIsBetter":false,"digits":1,"baseline":22.4,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":9.04,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":18.77,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":7.79,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":6.78,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"16e317c04d4a006c","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"nvidia/Cosmos-Reason2-8B","hardware":"thor","variant":"huggingface_fp16_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":15.3,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":6777,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":15.8,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":79.2,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"GSM8K","value":94.2,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"IFEval","value":79.3,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"MMLU-Pro","value":64.2,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":4.15,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.24,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":63.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":99.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":22.4,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":18.77,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"nvidia/Cosmos-Reason2-8B","hardware":"thor","variant":"huggingface_int4_embedl_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":37.8,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":15.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":3244,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":6777,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":46,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":15.8,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":77.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":79.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":93.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":94.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":78.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":79.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":61.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":64.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.89,"unit":"J","higherIsBetter":false,"digits":2,"baseline":4.15,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":0.53,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":0.24,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":71.4,"unit":"W","higherIsBetter":false,"digits":1,"baseline":63.7,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":91.7,"unit":"W","higherIsBetter":false,"digits":1,"baseline":99.5,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":23,"unit":"W","higherIsBetter":false,"digits":1,"baseline":22.4,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":9.04,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":18.77,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":7.79,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":6.78,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"16e317c04d4a006c","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"nvidia/Cosmos-Reason2-8B","hardware":"orin","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":25.1,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":2074,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":30.9,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":77.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":79.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":93.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":94.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":78.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":79.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":61.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":64.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.86,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.54,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":46.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":62.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":5.86,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":7.7,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":6.78,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"a74a64f3739235cc","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"nvidia/Cosmos-Reason2-8B","hardware":"orin","variant":"huggingface_int4_embedl_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":25.1,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":2074,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":30.9,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":77.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":79.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":93.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":94.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":78.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":79.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":61.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":64.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.86,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.54,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":46.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":62.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":5.86,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":7.7,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":6.78,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"a74a64f3739235cc","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"nvidia/cosmos3-edge","upstream":"nvidia/Cosmos3-Edge","title":"Cosmos3 Edge","source":"nvidia/Cosmos3-Edge","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=nvidia%2FCosmos3-Edge&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"nvidia/cosmos3-edge","hardware":"thor","precision":"INT4","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":200.5,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Decode","value":207.8,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":68.2,"baselineLabel":"BF16 on vLLM"}]},{"title":"Accuracy","subtitle":"MMMU-Pro, RealWorldQA, AI2D, MMBench, DocVQA · 6,765 questions · FP16 on TensorRT Edge-LLM as baseline","metrics":[{"description":"Average of 5","value":65.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.56,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMMU-Pro","value":24.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":24.87,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"RealWorldQA","value":70.59,"unit":"%","higherIsBetter":true,"digits":1,"baseline":69.28,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"AI2D","value":69.33,"unit":"%","higherIsBetter":true,"digits":1,"baseline":72.07,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"MMBench","value":82.07,"unit":"%","higherIsBetter":true,"digits":1,"baseline":81.47,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"DocVQA","value":79.62,"unit":"%","higherIsBetter":true,"digits":1,"baseline":80.13,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":2.13,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true}]},{"title":"Size","metrics":[{"description":"Engine size","value":1.42,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true}]}],"labels":["Image","Text","VLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"nvidia/cosmos3-edge","hardware":"orin-nx-16gb","precision":"INT4","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":69.6,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":3220,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":105.6,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"MMMU-Pro, RealWorldQA, AI2D, MMBench, DocVQA · 6,765 questions · FP16 on TensorRT Edge-LLM as baseline","metrics":[{"description":"Average of 5","value":65.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.56,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMMU-Pro","value":24.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":24.87,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"RealWorldQA","value":70.59,"unit":"%","higherIsBetter":true,"digits":1,"baseline":69.28,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"AI2D","value":69.33,"unit":"%","higherIsBetter":true,"digits":1,"baseline":72.07,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"MMBench","value":82.07,"unit":"%","higherIsBetter":true,"digits":1,"baseline":81.47,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"DocVQA","value":79.62,"unit":"%","higherIsBetter":true,"digits":1,"baseline":80.13,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.3073,"unit":"J","higherIsBetter":false,"digits":3,"headline":true},{"description":"Generated tokens per joule","value":3.255,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":21.388,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":22.647,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":8.851,"unit":"W","higherIsBetter":false,"digits":1}]},{"title":"Size","metrics":[{"description":"Container size, download","value":0.91,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.65,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"b6155c4ef041e021","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"nvidia/cosmos3-edge","hardware":"orin-nano-super","precision":"INT4","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":85.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":2619,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":152.5,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":26.8,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","subtitle":"MMMU-Pro, RealWorldQA, AI2D, MMBench, DocVQA · 6,765 questions · FP16 on TensorRT Edge-LLM as baseline","metrics":[{"description":"Average of 5","value":65.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.56,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMMU-Pro","value":24.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":24.87,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"RealWorldQA","value":70.59,"unit":"%","higherIsBetter":true,"digits":1,"baseline":69.28,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"AI2D","value":69.33,"unit":"%","higherIsBetter":true,"digits":1,"baseline":72.07,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"MMBench","value":82.07,"unit":"%","higherIsBetter":true,"digits":1,"baseline":81.47,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"DocVQA","value":79.62,"unit":"%","higherIsBetter":true,"digits":1,"baseline":80.13,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":3.1,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":5.03,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Size","metrics":[{"description":"Container size, download","value":0.91,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.65,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":2.46,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Engine size","value":1.26,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"b1ec138700741530","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/crossvit_15_dagger_240.in1k","title":"CrossViT-15 (240, dagger)","source":"timm/crossvit_15_dagger_240.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcrossvit_15_dagger_240.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/crossvit_15_dagger_240.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.04585,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.19366,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":946.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.34,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.5,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2271c82b289d90cf","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_15_dagger_240.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.49775,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.52252,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":398.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.34,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.5,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"859c0a3499c5877b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_15_dagger_240.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.55945,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.62067,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":218.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.34,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.5,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c4491edcd2690082","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_15_dagger_240.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.800974,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.97684,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.806213,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.98175,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1245.8,"unit":"qps","digits":1,"baseline":1021.89,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.34,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.5,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.84,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.32,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"26b1f2d8dba93c5b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_15_dagger_240.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.34329,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.36194,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":743.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.34,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.5,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"13fdf28898ca5614","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/crossvit_15_240.in1k","title":"CrossViT-15 (240)","source":"timm/crossvit_15_240.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcrossvit_15_240.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/crossvit_15_240.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.36307,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.49854,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":725.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.2,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.55,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"574a88aafd80d7c1","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_15_240.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.46902,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.49854,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":402.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.2,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.55,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1bf08ba75ee652a7","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_15_240.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.48339,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.53503,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":221.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.2,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.55,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5f6ab50833bc6a9e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_15_240.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.800123,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.888863,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.80542,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.896545,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1247.2,"unit":"qps","digits":1,"baseline":1123.26,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.2,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.55,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.83,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.31,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"af9358f63c6e5f16","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_15_240.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.30233,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.32098,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":767.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.2,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.55,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"248a16e4b220b936","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/crossvit_15_dagger_408.in1k","title":"CrossViT-15 (408, dagger)","source":"timm/crossvit_15_dagger_408.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcrossvit_15_dagger_408.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/crossvit_15_dagger_408.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.66108,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.78101,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":591.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.09,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.35,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"366132b6fd0200bf","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_15_dagger_408.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.88201,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.91504,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":256.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.09,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.35,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ba91533379b29ea2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_15_dagger_408.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.2379,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":8.29712,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":120.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.09,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.35,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a6e8ba457ff52ff6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_15_dagger_408.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.31222,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.45886,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":231.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.09,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.35,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.85,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.33,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"dfdcadcb6b7ec92a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_15_dagger_408.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.24557,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.30042,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":235.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.09,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.35,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a104dcb498327a5b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/crossvit_18_dagger_240.in1k","title":"CrossViT-18 (240, dagger)","source":"timm/crossvit_18_dagger_240.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcrossvit_18_dagger_240.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/crossvit_18_dagger_240.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.38337,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.51154,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":715.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.07,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.84,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"72ededd257a8583a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_18_dagger_240.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.99744,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.0293,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":331.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.07,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.84,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7d2dca0e06a8001c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_18_dagger_240.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.62414,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.69592,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":176.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.07,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.84,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c60d098bdb3a44aa","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_18_dagger_240.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.06987,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.1851,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.07526,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.19141,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":933.5,"unit":"qps","digits":1,"baseline":842.824,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.07,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.84,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.02,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.5,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"39b50c0302c3e6b2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_18_dagger_240.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.64992,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.67279,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":605.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.07,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.84,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1695e8bea207fa31","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/crossvit_18_240.in1k","title":"CrossViT-18 (240)","source":"timm/crossvit_18_240.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcrossvit_18_240.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/crossvit_18_240.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.64786,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.77075,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":598.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.83,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d6964f217d32a9ce","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_18_240.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.95832,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.00049,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":336.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.83,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1519c7fd6d2764d2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_18_240.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.53371,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.61456,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":179.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.83,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"05f5ac1b2a961f8d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_18_240.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.04588,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.202,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.0498,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.20813,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":954.9,"unit":"qps","digits":1,"baseline":830.995,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.83,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.49,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"0d4ec6904be1a0d2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_18_240.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.59975,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.61737,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":624.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.83,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2ff283d0dcac2aa2","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/crossvit_18_dagger_408.in1k","title":"CrossViT-18 (408, dagger)","source":"timm/crossvit_18_dagger_408.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcrossvit_18_dagger_408.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/crossvit_18_dagger_408.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.13576,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.2572,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":462.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.66,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2ade727fe5e564b4","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_18_dagger_408.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.05803,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.94953,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":5.07819,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.97144,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":196.7,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":167.205,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.66,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"00611d371603aa93","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_18_dagger_408.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":11.5714,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":11.636,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":86,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.66,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"fe2aa731f6b86dc8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_18_dagger_408.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.66907,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.88843,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":176.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.66,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.04,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.51,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"db2b99dc3df22a05","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_18_dagger_408.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.64669,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.68225,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":177.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.66,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e171b302824abefd","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/crossvit_base_240.in1k","title":"CrossViT-240 Base","source":"timm/crossvit_base_240.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcrossvit_base_240.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/crossvit_base_240.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.89265,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.00934,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":521.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.76,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"de6e20668cec0271","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_base_240.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.59823,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.15206,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":3.61353,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.17212,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":276.4,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":239.567,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.76,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"29d3630f2ab438f5","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_base_240.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.42761,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":7.45264,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":133.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.76,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4e487f853d687c60","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_base_240.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.38697,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.85332,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.39392,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.9248,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":720.3,"unit":"qps","digits":1,"baseline":539.146,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.76,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.7,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.18,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"bae6035606ebcd4d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_base_240.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.96707,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.97772,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":508.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.76,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8c2e16a61d6ce936","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/crossvit_small_240.in1k","title":"CrossViT-240 Small","source":"timm/crossvit_small_240.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcrossvit_small_240.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/crossvit_small_240.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.30286,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.43298,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":758.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.28,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.18,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9ee749b15bbf327f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_small_240.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.19503,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.23071,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":453,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.28,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.18,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7c3086a6943f3193","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_small_240.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.05892,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.13214,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":245.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.28,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.18,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3b1de59facc0a087","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_small_240.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.69965,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.877983,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.705963,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.885132,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1426.1,"unit":"qps","digits":1,"baseline":1137.11,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.28,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.18,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.83,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.3,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"7cc04bab38bf6b95","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_small_240.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.20566,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.22528,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":828.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.28,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.18,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"42b0d0d6b31c7951","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/crossvit_tiny_240.in1k","title":"CrossViT-240 Tiny","source":"timm/crossvit_tiny_240.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcrossvit_tiny_240.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/crossvit_tiny_240.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.985673,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.12775,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":997.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.5,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.37,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"65162f382311bfd9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_tiny_240.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.63106,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.65698,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":609.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.5,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.37,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ade5cafba67c98dc","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_tiny_240.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.5751,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.60901,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":386.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.5,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.37,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"89adad58b94a586a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_tiny_240.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.524285,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.525696,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1902.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":72.5,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.37,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.6,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":82.9,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"048608de93eb1ae6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_tiny_240.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.907912,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.913452,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1100.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.5,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.37,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4cb5e8b6e621ce45","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/crossvit_9_dagger_240.in1k","title":"CrossViT-9 (240, dagger)","source":"timm/crossvit_9_dagger_240.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcrossvit_9_dagger_240.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/crossvit_9_dagger_240.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.739831,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.761353,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1292.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.3,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.64,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7104afe3b608703e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_9_dagger_240.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.55436,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.57526,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":639.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.3,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.64,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"578ec545a1b5035f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_9_dagger_240.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.48041,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.52698,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":400.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.3,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.64,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ed2fd12eddcc64e0","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_9_dagger_240.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.478908,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.481384,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2082.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":75.3,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.64,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.62,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.1,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"12aa04c8e9818b79","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_9_dagger_240.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.84539,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.851929,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1181.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.3,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.64,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f897acebc7af5a5f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/crossvit_9_240.in1k","title":"CrossViT-9 (240)","source":"timm/crossvit_9_240.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcrossvit_9_240.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/crossvit_9_240.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.958488,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.06781,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1031.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.11,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0e2e7aa7e133fee2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_9_240.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.53023,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.56256,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":649.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.11,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9767984803b10029","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_9_240.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.4001,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.44702,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":414.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.11,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8fe4581e4f2e0b6e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_9_240.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.485169,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.488831,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2054.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":72.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.11,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.62,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":100.5,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"bd3f7a05702c1ee2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/crossvit_9_240.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.825603,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.834229,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1209.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.11,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0cb284237b05713e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/cs3darknet_focus_m.c2ns_in1k","title":"CS3-Darknet (focus, m)","source":"timm/cs3darknet_focus_m.c2ns_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcs3darknet_focus_m.c2ns_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/cs3darknet_focus_m.c2ns_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.54457,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.546265,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1831.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.38,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.37,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.64,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.11,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"2f0309fe375ae58c","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/cs3darknet_focus_s.ra4_e3600_r256_in1k","title":"CS3-Darknet (focus, s)","source":"timm/cs3darknet_focus_s.ra4_e3600_r256_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcs3darknet_focus_s.ra4_e3600_r256_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/cs3darknet_focus_s.ra4_e3600_r256_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.311191,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.312439,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3195.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":72.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.76,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.56,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.04,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"49e7aa57fc4e00f3","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/cs3darknet_l.c2ns_in1k","title":"CS3-Darknet-L","source":"timm/cs3darknet_l.c2ns_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcs3darknet_l.c2ns_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/cs3darknet_l.c2ns_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.580612,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.58197,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1718.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.89,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.6,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.76,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.24,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"882fd40385ac9e9a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/cs3darknet_m.c2ns_in1k","title":"CS3-Darknet-M","source":"timm/cs3darknet_m.c2ns_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcs3darknet_m.c2ns_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/cs3darknet_m.c2ns_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.560826,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.563904,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1778.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.89,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.52,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.64,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.11,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"3827be110df9b9c3","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/cspdarknet53.ra_in1k","title":"CSPDarknet-53","source":"timm/cspdarknet53.ra_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcspdarknet53.ra_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/cspdarknet53.ra_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.58024,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.771763,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.649414,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.858154,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1677.2,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1259.22,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.47,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e852a3881057d348","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cspdarknet53.ra_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.4396,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.66765,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1.44653,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.67508,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":690.5,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":596.115,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.47,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ea5a91295027c92f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cspdarknet53.ra_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.78687,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.81494,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":356.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.47,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2b43515d4b261cae","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cspdarknet53.ra_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.774887,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.778198,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1288,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.47,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.83,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.31,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"d005cd89d3864667","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cspdarknet53.ra_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.30819,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.31244,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":763.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.47,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f0322a29ceefd653","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/cspresnet50.ra_in1k","title":"CSPResNet-50","source":"timm/cspresnet50.ra_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcspresnet50.ra_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/cspresnet50.ra_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.618372,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.684937,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1555,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.48,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.69,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6921a3cc32a378aa","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cspresnet50.ra_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.32916,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.33704,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":747.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.48,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.69,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9a736ee24a89ea06","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cspresnet50.ra_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.55977,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.27075,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":388.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.48,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.69,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b3dbb31e86dcef31","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cspresnet50.ra_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.49082,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.49585,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":670,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.48,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.69,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.84,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.32,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"205011b40284f5ac","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cspresnet50.ra_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.63093,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.64349,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":612.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.48,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.69,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ae4d91cf1ca2f603","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/cspresnext50.ra_in1k","title":"CSPResNeXt-50","source":"timm/cspresnext50.ra_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcspresnext50.ra_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/cspresnext50.ra_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.744821,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.811218,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1295.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.58,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.31,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9be0a1041481f1c4","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cspresnext50.ra_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.45639,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.46332,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":682.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.58,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.31,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1eed9920ab89bc36","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cspresnext50.ra_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.95917,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.97394,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":336.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.58,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.31,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"75d101e05228c169","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cspresnext50.ra_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.23713,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.24225,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":807.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.58,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.31,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.81,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.29,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"ccfb6ba73e81a62f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/cspresnext50.ra_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.37219,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.37848,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":728.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.58,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.31,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"cbb6f47d46c01ec4","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/darknet53.c2ns_in1k","title":"Darknet-53","source":"timm/darknet53.c2ns_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdarknet53.c2ns_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/darknet53.c2ns_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.899271,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.90274,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1109.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.17,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.72,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.47,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"1e884d8805c504ea","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/darknetaa53.c2ns_in1k","title":"Darknet-AA-53","source":"timm/darknetaa53.c2ns_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdarknetaa53.c2ns_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/darknetaa53.c2ns_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.852792,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.855469,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1170.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.72,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.51,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.4,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"2beccfd19316ef0f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/davit_base_fl.msft_florence2","title":"DaViT Base (fl)","source":"timm/davit_base_fl.msft_florence2","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdavit_base_fl.msft_florence2&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/davit_base_fl.msft_florence2","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":17.17,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":17.843,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":58.2,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.32,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.8,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"3ee9295032d03115","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/davit_huge_fl.msft_florence2","title":"DaViT Huge (fl)","source":"timm/davit_huge_fl.msft_florence2","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdavit_huge_fl.msft_florence2&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/davit_huge_fl.msft_florence2","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":48.2671,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":49.5062,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":20.7,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.67,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.15,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"6b243853a95a0b96","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/davit_tiny.msft_in1k","title":"DaViT Tiny","source":"timm/davit_tiny.msft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdavit_tiny.msft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/davit_tiny.msft_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.33686,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.44513,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":733.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.19,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.02,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a10a17799207f4b3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/davit_tiny.msft_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.38589,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.3952,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":416.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.19,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.02,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5dfc66b48fe2892f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/davit_tiny.msft_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.84197,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.85461,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":205.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.19,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.02,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e100c8bca6767cd7","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/davit_tiny.msft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.05987,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.06311,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":942.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.19,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.02,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.85,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.33,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"256f5cda2a42d9c0","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/davit_tiny.msft_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.37524,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.37872,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":726.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.19,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.02,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5411b8d413791dbe","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/deeplabv3_mobilenet_v3_large","title":"DeepLabV3 MobileNetV3 Large","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fdeeplabv3_mobilenet_v3_large&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/deeplabv3_mobilenet_v3_large","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.843,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.85,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":542.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"COCO segmentation validation · 2,000 images · Baseline TensorRT as baseline","metrics":[{"description":"mIoU","value":45.68,"unit":"%","headline":true}]}],"labels":["Image","Segmentation"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/deeplabv3_mobilenet_v3_large","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.843,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.85,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":542.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"COCO segmentation validation · 2,000 images · Baseline TensorRT as baseline","metrics":[{"description":"mIoU","value":45.68,"unit":"%","headline":true}]}],"labels":["Image","Segmentation"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/deeplabv3_resnet101","title":"DeepLabV3 ResNet-101","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fdeeplabv3_resnet101&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/deeplabv3_resnet101","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":14.536,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":15.222,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":68.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"COCO segmentation validation · 2,000 images · Baseline TensorRT as baseline","metrics":[{"description":"mIoU","value":67.06,"unit":"%","headline":true}]}],"labels":["Image","Segmentation"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/deeplabv3_resnet101","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":14.536,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":15.222,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":68.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"COCO segmentation validation · 2,000 images · Baseline TensorRT as baseline","metrics":[{"description":"mIoU","value":67.06,"unit":"%","headline":true}]}],"labels":["Image","Segmentation"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/deeplabv3_resnet50","title":"DeepLabV3 ResNet-50","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fdeeplabv3_resnet50&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/deeplabv3_resnet50","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":8.855,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":9.217,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":112.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"COCO segmentation validation · 2,000 images · Baseline TensorRT as baseline","metrics":[{"description":"mIoU","value":64.93,"unit":"%","headline":true}]}],"labels":["Image","Segmentation"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/deeplabv3_resnet50","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":8.855,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":9.217,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":112.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"COCO segmentation validation · 2,000 images · Baseline TensorRT as baseline","metrics":[{"description":"mIoU","value":64.93,"unit":"%","headline":true}]}],"labels":["Image","Segmentation"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/deit_base_distilled_patch16_224.fb_in1k","title":"DeiT Base/16 (224, distilled)","source":"timm/deit_base_distilled_patch16_224.fb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdeit_base_distilled_patch16_224.fb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/deit_base_distilled_patch16_224.fb_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.05249,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.21271,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1.12341,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.28296,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":931.3,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":806.038,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.1,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1b076f2671afc023","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_base_distilled_patch16_224.fb_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.46566,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.97633,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":2.47766,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.99066,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":403.3,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":334.137,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.1,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"38bca917b4c50b41","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_base_distilled_patch16_224.fb_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.12542,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.20105,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":194.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.1,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1a8a0228d73c6386","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_base_distilled_patch16_224.fb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.993771,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.29208,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.99707,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.29596,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1004.8,"unit":"qps","digits":1,"baseline":772.856,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.1,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.5,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.98,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"b2d3930775e75af2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_base_distilled_patch16_224.fb_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.37,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.37482,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":729.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.1,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"dedc9493fd0ac52a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/deit_base_patch16_224.fb_in1k","title":"DeiT Base/16 (224)","source":"timm/deit_base_patch16_224.fb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdeit_base_patch16_224.fb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/deit_base_patch16_224.fb_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.0275,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.19549,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1.11401,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.26459,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":949.6,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":818.664,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.3,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.33,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"26e9a1ea0c31a121","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_base_patch16_224.fb_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.36945,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.98304,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":2.38342,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.99927,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":419.7,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":333.391,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.3,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.33,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a44772cf0fe680d6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_base_patch16_224.fb_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.94205,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.96228,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":201.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.3,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.33,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d840e584b1f80bc7","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_base_patch16_224.fb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.982204,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.28756,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.987427,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.30035,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1016.5,"unit":"qps","digits":1,"baseline":775.679,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.3,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.33,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.49,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.97,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"26ca067fe9134595","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_base_patch16_224.fb_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.36146,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.3656,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":734,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.3,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.33,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"aa46d9fc7fdff866","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/deit_base_distilled_patch16_384.fb_in1k","title":"DeiT Base/16 (384, distilled)","source":"timm/deit_base_distilled_patch16_384.fb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdeit_base_distilled_patch16_384.fb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/deit_base_distilled_patch16_384.fb_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.64092,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.71954,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":600.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.24,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2abd31a626b3fd90","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_base_distilled_patch16_384.fb_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.56449,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.41038,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":4.57227,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.42603,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":217.9,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":183.86,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.24,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4403321a03707300","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_base_distilled_patch16_384.fb_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":11.7251,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":11.7446,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":84.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.24,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"921547783a137ac5","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_base_distilled_patch16_384.fb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.28119,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.56488,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":189.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.24,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.52,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"1b4e016f6a4ee014","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_base_distilled_patch16_384.fb_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.7693,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.83899,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":209.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.24,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d0066506e0e403b1","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/deit_base_patch16_384.fb_in1k","title":"DeiT Base/16 (384)","source":"timm/deit_base_patch16_384.fb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdeit_base_patch16_384.fb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/deit_base_patch16_384.fb_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.64139,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.72473,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":600.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.31,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"bff09fff9530b8b5","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_base_patch16_384.fb_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.4527,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.32474,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":4.46405,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.34985,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":223.4,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":186.817,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.31,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"69081b674c3ec1d2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_base_patch16_384.fb_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":11.3098,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":11.3317,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":88,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.31,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"064b9f38add463e3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_base_patch16_384.fb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.04038,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.27039,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":198.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.31,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"cbab905117422afe","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_base_patch16_384.fb_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.78216,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.80499,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":209.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.31,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"39adc5b4af216691","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/deit3_base_patch16_224.fb_in1k","title":"DeiT III Base/16 (224)","source":"timm/deit3_base_patch16_224.fb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdeit3_base_patch16_224.fb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/deit3_base_patch16_224.fb_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.04982,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.20484,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1.13614,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.27313,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":932.6,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":811.804,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.38,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f65b2c78db10a357","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_base_patch16_224.fb_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.42479,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.88917,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":2.43823,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.92346,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":410.1,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":344.227,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.38,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3c392a95e9721e30","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_base_patch16_224.fb_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.94701,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.00018,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":201.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.38,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"05a1e8c06d4158ab","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_base_patch16_224.fb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.989849,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.29498,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.992126,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.29791,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1008.6,"unit":"qps","digits":1,"baseline":771.237,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.38,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.49,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.97,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"fb28d71b3b819da4","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_base_patch16_224.fb_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.35665,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.36188,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":736.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.38,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"02160147270060c1","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/deit3_base_patch16_384.fb_in1k","title":"DeiT III Base/16 (384)","source":"timm/deit3_base_patch16_384.fb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdeit3_base_patch16_384.fb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/deit3_base_patch16_384.fb_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.64849,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.73389,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":597.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.08,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f7e21cb57d3872e6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_base_patch16_384.fb_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.4237,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.33985,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":4.4364,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.35083,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":224.9,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":186.29,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.08,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3c0c4bdcfe27b959","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_base_patch16_384.fb_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":11.4116,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":11.4412,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":87.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.08,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"968fceea3eabc5e9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_base_patch16_384.fb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.28895,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.61969,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":189,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.08,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.52,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"3a06445900ae9f8c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_base_patch16_384.fb_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.78289,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.80884,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":209,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.08,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"88cc490b1a81696d","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/deit3_huge_patch14_224.fb_in1k","title":"DeiT III Huge/14 (224)","source":"timm/deit3_huge_patch14_224.fb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdeit3_huge_patch14_224.fb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/deit3_huge_patch14_224.fb_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.01647,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":7.21778,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":5.16333,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":7.39368,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":197.6,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":137.461,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.96,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.18,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"dbafb18473d474fc","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_huge_patch14_224.fb_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":13.7204,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":17.1831,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":13.7502,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":17.2314,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":72.5,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":57.9019,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.96,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.18,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b5f0fff5017a32d2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_huge_patch14_224.fb_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":33.9987,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":34.3119,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":29.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.96,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.18,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"de0954ed74a2e01e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_huge_patch14_224.fb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.29807,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":9.59521,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":107.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.96,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.18,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":7.75,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":7.23,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"38ba4a5ec0d0b6a9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_huge_patch14_224.fb_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":11.9072,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":12.0278,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":84,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.96,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.18,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"19ba6b5edeb743a5","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/deit3_large_patch16_224.fb_in1k","title":"DeiT III Large/16 (224)","source":"timm/deit3_large_patch16_224.fb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdeit3_large_patch16_224.fb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/deit3_large_patch16_224.fb_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.53656,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.38278,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":2.63812,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.48859,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":390.6,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":292.336,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.91,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"bc19359146974895","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_large_patch16_224.fb_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.38523,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":8.11363,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":6.401,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":8.12927,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":155.8,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":122.615,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.91,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"59d5efdb7114154e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_large_patch16_224.fb_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":13.5212,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":13.5447,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":73.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.91,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"adf5c0034c0c2022","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_large_patch16_224.fb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.14967,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.43336,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":3.21576,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.66821,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":317.3,"unit":"qps","digits":1,"baseline":225.477,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.91,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.93,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.41,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"d5d76b0c17b807c1","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_large_patch16_224.fb_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.18179,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.20325,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":314.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.91,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"421a393577620c56","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/deit3_large_patch16_384.fb_in1k","title":"DeiT III Large/16 (384)","source":"timm/deit3_large_patch16_384.fb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdeit3_large_patch16_384.fb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/deit3_large_patch16_384.fb_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.22635,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.31128,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":234.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.67,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.58,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"eebef7d226a32df6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_large_patch16_384.fb_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":13.6305,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":15.7788,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":13.6526,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":15.8098,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":73,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":63.0548,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.67,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.58,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"467e39ff0b87c7ed","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_large_patch16_384.fb_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":33.1162,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":33.4443,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":30,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.67,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.58,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"79b7018322b32e39","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_large_patch16_384.fb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":14.2443,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":14.8854,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":70.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":85.67,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.58,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.48,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"9f82f29b3d9acf42","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_large_patch16_384.fb_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":14.1164,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":14.2219,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":70.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.67,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.58,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b3d63a21b6c46fbe","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/deit3_medium_patch16_224.fb_in1k","title":"DeiT III Medium/16 (224)","source":"timm/deit3_medium_patch16_224.fb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdeit3_medium_patch16_224.fb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/deit3_medium_patch16_224.fb_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.798643,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.87915,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1203.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.38,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.05,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ff52fa76cc481b78","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_medium_patch16_224.fb_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.52966,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.68721,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1.53552,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.70099,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":649.9,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":589.226,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.38,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.05,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b42d7041f2b2840f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_medium_patch16_224.fb_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.06907,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.10876,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":324,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.38,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.05,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"19756a614bfe3f41","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_medium_patch16_224.fb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.589109,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.800436,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.592651,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.802551,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1692.9,"unit":"qps","digits":1,"baseline":1246.86,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.38,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.05,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.96,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.44,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"5a6236db62f39538","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_medium_patch16_224.fb_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.851235,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.854004,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1173.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.38,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.05,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"776e27077e4561e1","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/deit3_small_patch16_224.fb_in1k","title":"DeiT III Small/16 (224)","source":"timm/deit3_small_patch16_224.fb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdeit3_small_patch16_224.fb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/deit3_small_patch16_224.fb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.472609,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.473938,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2109.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.44,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.77,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.77,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.25,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"6ce2ee5c70883486","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/deit3_small_patch16_384.fb_in1k","title":"DeiT III Small/16 (384)","source":"timm/deit3_small_patch16_384.fb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdeit3_small_patch16_384.fb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/deit3_small_patch16_384.fb_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.02796,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.09802,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":951.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.7,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.38,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1e41970632505087","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_small_patch16_384.fb_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.09155,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.09747,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":475.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.7,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.38,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4e194444fc30847a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_small_patch16_384.fb_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.77106,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.78149,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":208.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.7,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.38,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6d7bd41f3a685483","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_small_patch16_384.fb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.11319,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.21838,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":472.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.7,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.38,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.78,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.26,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e1a2fe4077596e52","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit3_small_patch16_384.fb_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.27926,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.29138,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":438.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.7,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.38,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"966a79d3252f6048","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/deit_small_distilled_patch16_224.fb_in1k","title":"DeiT Small/16 (224, distilled)","source":"timm/deit_small_distilled_patch16_224.fb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdeit_small_distilled_patch16_224.fb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/deit_small_distilled_patch16_224.fb_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.69692,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.769714,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1390.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.12,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"17104a8b232662f9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_small_distilled_patch16_224.fb_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.3284,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.33411,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":748.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.12,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c42c0142786f85c7","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_small_distilled_patch16_224.fb_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.50495,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.53571,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":396.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.12,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c1979288c96c919e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_small_distilled_patch16_224.fb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.476813,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.478088,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2089.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.12,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.77,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.25,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"6ef59fd9d2fa61ef","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_small_distilled_patch16_224.fb_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.699539,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.704346,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1427.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.12,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"54ebd75acaca0593","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/deit_small_patch16_224.fb_in1k","title":"DeiT Small/16 (224)","source":"timm/deit_small_patch16_224.fb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdeit_small_patch16_224.fb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/deit_small_patch16_224.fb_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.686766,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.759827,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1411,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.66,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4382f5bf82f8c036","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_small_patch16_224.fb_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.26743,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.27277,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":784.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.66,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ffaa5db496e232eb","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_small_patch16_224.fb_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.47898,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.50537,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":401.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.66,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d73ab3fe807810cc","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_small_patch16_224.fb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.464932,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.516376,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.465881,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.517822,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":2143.6,"unit":"qps","digits":1,"baseline":1930.7,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.66,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.77,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.25,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"b7359113c6e284af","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_small_patch16_224.fb_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.693132,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.695984,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1440.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.66,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"66e52c324097cdc6","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/deit_tiny_distilled_patch16_224.fb_in1k","title":"DeiT Tiny/16 (224, distilled)","source":"timm/deit_tiny_distilled_patch16_224.fb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdeit_tiny_distilled_patch16_224.fb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/deit_tiny_distilled_patch16_224.fb_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.433301,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.523499,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2256,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":73.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.28,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8da69101577b326d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_tiny_distilled_patch16_224.fb_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.905325,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.910278,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1097.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":73.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.28,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"bbb83a4a5244e326","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_tiny_distilled_patch16_224.fb_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.50603,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.51379,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":660,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":73.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.28,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a64d1cb6d73c8910","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_tiny_distilled_patch16_224.fb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.347445,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.347961,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2867,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":73.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.28,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":69.1,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"63ad82abee72ec8a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_tiny_distilled_patch16_224.fb_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.505055,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.507568,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1975.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":73.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.28,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"cabfc9eff08ad12c","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/deit_tiny_patch16_224.fb_in1k","title":"DeiT Tiny/16 (224)","source":"timm/deit_tiny_patch16_224.fb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdeit_tiny_patch16_224.fb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/deit_tiny_patch16_224.fb_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.429432,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.519531,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2272.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":71.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.41,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2bf41209da3d9a73","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_tiny_patch16_224.fb_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.871305,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.875671,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1140.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":71.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.41,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e817df3ca4d1b9e8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_tiny_patch16_224.fb_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.48133,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.48773,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":671,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":71.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.41,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4eb48ad1c42bc276","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_tiny_patch16_224.fb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.343965,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.34436,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2894,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":71.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.41,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":66.9,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"341e322dc0624c0c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/deit_tiny_patch16_224.fb_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.500849,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.501892,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1992.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":71.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.41,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c5790fa907dd1a91","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"facebook/deit-base-patch16-224","title":"DeiT-Base","source":"facebook/deit-base-patch16-224","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=facebook%2Fdeit-base-patch16-224&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"facebook/deit-base-patch16-224","hardware":"thor","precision":"FP8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.06421,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.39966,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Throughput","value":932.835,"unit":"qps","digits":1,"baseline":706.353,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNette validation · 3,925 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":92.35668789808918,"unit":"%","digits":2,"baseline":92,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Top-5","value":99.61783439490446,"unit":"%","digits":2,"baseline":99.56687898089172,"baselineLabel":"Baseline TensorRT"}]},{"title":"Memory","metrics":[{"description":"Peak device memory","value":86,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":167,"baselineLabel":"Baseline TensorRT","improvementWord":"smaller","headline":true}]},{"title":"Size","metrics":[{"description":"Engine size","value":85.9,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":167.2,"baselineLabel":"Baseline TensorRT","improvementWord":"smaller","headline":true}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"facebook/deit-base-patch16-224","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.0813,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.39417,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Throughput","value":922.426,"unit":"qps","digits":1,"baseline":719.884,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNette validation · 3,925 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":91.6687898089172,"unit":"%","digits":2,"baseline":91.94904458598727,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Top-5","value":99.56687898089172,"unit":"%","digits":2,"baseline":99.56687898089172,"baselineLabel":"Baseline TensorRT"}]},{"title":"Memory","metrics":[{"description":"Peak device memory","value":92,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":168,"baselineLabel":"Baseline TensorRT","improvementWord":"smaller","headline":true}]},{"title":"Size","metrics":[{"description":"Engine size","value":94.5,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":168.1,"baselineLabel":"Baseline TensorRT","improvementWord":"smaller","headline":true}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"facebook/deit-tiny-patch16-224","title":"DeiT-Tiny","source":"facebook/deit-tiny-patch16-224","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=facebook%2Fdeit-tiny-patch16-224&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"facebook/deit-tiny-patch16-224","hardware":"thor","precision":"FP8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.456421,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.459717,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Throughput","value":2179.8,"unit":"qps","digits":1,"baseline":2166.29,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNette validation · 3,925 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":70.19108280254777,"unit":"%","digits":2,"baseline":70.62420382165605,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Top-5","value":91.26114649681529,"unit":"%","digits":2,"baseline":91.4140127388535,"baselineLabel":"Baseline TensorRT"}]},{"title":"Memory","metrics":[{"description":"Peak device memory","value":8,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":11,"baselineLabel":"Baseline TensorRT","improvementWord":"smaller","headline":true}]},{"title":"Size","metrics":[{"description":"Engine size","value":9.3,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":12.7,"baselineLabel":"Baseline TensorRT","improvementWord":"smaller","headline":true}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"torchvision/densenet121","title":"DenseNet-121","source":"torchvision/densenet121","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fdensenet121&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/densenet121","hardware":"orin-nano-super","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":7.28394,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":137.111,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNette validation · 3,925 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":79.36305732484077,"unit":"%","digits":2,"baseline":79.51592356687898,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":96.02547770700637,"unit":"%","digits":2,"baseline":96.35668789808918,"baselineLabel":"FP32 model"}]},{"title":"Size","metrics":[{"description":"Engine size","value":10,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"torchvision/densenet121","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":2.515079639852047,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":397.6017236809354,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":74.156,"unit":"%","digits":2,"baseline":74.434,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":91.81200000000001,"unit":"%","digits":2,"baseline":91.972,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/densenet121.ra_in1k","title":"DenseNet-121 (timm)","source":"timm/densenet121.ra_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdensenet121.ra_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/densenet121.ra_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.788077,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.892761,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1257.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.94,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.34,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2fe5d9530a33194a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/densenet121.ra_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.28771,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.37366,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":186.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.94,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.34,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1614967c42bdd467","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/densenet121.ra_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.53791,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":7.58936,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":132.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.94,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.34,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0454ee2fe6520492","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/densenet121.ra_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.61586,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.6178,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":618.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":74.94,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.34,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.61,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":92.1,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"b73e9411d8641cf9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/densenet121.ra_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.90183,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.90424,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":344.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.94,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.34,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ffb3282e2659248f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/densenet161","title":"DenseNet-161","source":"torchvision/densenet161","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdensenet161.tv_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/densenet161","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":4.631678340956569,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":215.90445760390878,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":77.046,"unit":"%","digits":2,"baseline":77.138,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":93.45400000000001,"unit":"%","digits":2,"baseline":93.56,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/densenet161.tv_in1k","title":"DenseNet-161 (timm)","source":"timm/densenet161.tv_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdensenet161.tv_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/densenet161.tv_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.39741,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.52332,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":708,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.22,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.59,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7145b7c9f6b5b872","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/densenet161.tv_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":10.0703,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":10.181,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":98.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.22,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.59,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"08a22e68826dad43","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/densenet161.tv_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":14.5181,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":14.5437,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":68.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.22,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.59,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f592d76314a3006b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/densenet161.tv_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.00423,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.00702,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":332.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.22,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.59,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.85,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.33,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e0686afe09b79e28","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/densenet161.tv_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.5599,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.5744,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":179.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.22,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.59,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5f7115dcf652a294","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/densenet169","title":"DenseNet-169","source":"torchvision/densenet169","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdensenet169.tv_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/densenet169","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":5.019917440076824,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":199.20646344030234,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":75.388,"unit":"%","digits":2,"baseline":75.6,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":92.62599999999999,"unit":"%","digits":2,"baseline":92.806,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/densenet169.tv_in1k","title":"DenseNet-169 (timm)","source":"timm/densenet169.tv_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdensenet169.tv_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/densenet169.tv_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.05684,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.20807,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":937.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.84,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9ec1847559f8461a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/densenet169.tv_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":10.3166,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":10.4292,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":96.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.84,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7be5f22989ffe68d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/densenet169.tv_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":12.4951,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":13.0472,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":79.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.84,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"81e78fec35e74cd6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/densenet169.tv_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.82075,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.82245,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":354.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":75.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.84,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.68,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.16,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"c041e9560c53eb82","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/densenet169.tv_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.60135,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.60718,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":217.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.84,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b63067ca9f729991","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/densenet201","title":"DenseNet-201","source":"torchvision/densenet201","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdensenet201.tv_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/densenet201","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":7.535080220550299,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":132.71258841713677,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":76.658,"unit":"%","digits":2,"baseline":76.896,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":93.16799999999999,"unit":"%","digits":2,"baseline":93.37,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/densenet201.tv_in1k","title":"DenseNet-201 (timm)","source":"timm/densenet201.tv_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdensenet201.tv_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/densenet201.tv_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.27656,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.52794,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1.42584,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.67957,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":778.3,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":647.856,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.97,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.33,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3c05cee8f2e0431e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/densenet201.tv_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":15.308,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":15.782,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":65,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.97,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.33,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"fcaa4271ae08e938","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/densenet201.tv_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":17.9876,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":18.1009,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":55.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.97,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.33,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5f4362ef63a470c6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/densenet201.tv_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.02286,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.02637,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":248.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.97,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.33,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.75,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.23,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"f6f1caf7852314c8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/densenet201.tv_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.53676,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.54315,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":153,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.97,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.33,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"546747f54118f515","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"depth-anything/Depth-Anything-V2-Small-hf","upstream":"depth-anything/Depth-Anything-V2-Small-hf","title":"Depth Anything V2 Small","upstreamUrl":"https://huggingface.co/depth-anything/Depth-Anything-V2-Small-hf","source":"depth-anything/Depth-Anything-V2-Small-hf","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=depth-anything%2FDepth-Anything-V2-Small-hf&granularity=auto","builds":[{"model":"depth-anything/Depth-Anything-V2-Small-hf","hardware":"thor","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":2.512831926345825,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true}]},{"title":"Accuracy","subtitle":"Hypersim","metrics":[{"description":"AbsRel","value":0.06057747843652344,"higherIsBetter":false,"digits":4,"headline":true},{"description":"δ1","value":96.89650632021778,"unit":"%","higherIsBetter":true,"digits":2}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":532.7,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":50.71,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true},{"description":"Container size","value":703,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image","Depth"],"build":"20261018202343e9","updatedAt":"2026-11-04T20:59:53.000Z"},{"model":"depth-anything/Depth-Anything-V2-Small-hf","hardware":"thor","variant":"huggingface_fp16_tensorrt","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":2.512831926345825,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true}]},{"title":"Accuracy","subtitle":"Hypersim","metrics":[{"description":"AbsRel","value":0.06057747843652344,"higherIsBetter":false,"digits":4,"headline":true},{"description":"δ1","value":96.89650632021778,"unit":"%","higherIsBetter":true,"digits":2}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":532.7,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":50.71,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true},{"description":"Container size","value":703,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image","Depth"],"build":"20261018202343e9","updatedAt":"2026-11-04T20:59:53.000Z"},{"model":"depth-anything/Depth-Anything-V2-Small-hf","hardware":"orin","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":7.923007965087891,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true}]},{"title":"Accuracy","subtitle":"Hypersim","metrics":[{"description":"AbsRel","value":0.0606292045789222,"higherIsBetter":false,"digits":4,"headline":true},{"description":"δ1","value":96.89556147257557,"unit":"%","higherIsBetter":true,"digits":2}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":549.5,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":50.52,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image","Depth"],"build":"202611032015107e","updatedAt":"2026-11-04T20:59:53.000Z"},{"model":"depth-anything/Depth-Anything-V2-Small-hf","hardware":"orin","variant":"huggingface_fp16_tensorrt","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":7.923007965087891,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true}]},{"title":"Accuracy","subtitle":"Hypersim","metrics":[{"description":"AbsRel","value":0.0606292045789222,"higherIsBetter":false,"digits":4,"headline":true},{"description":"δ1","value":96.89556147257557,"unit":"%","higherIsBetter":true,"digits":2}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":549.5,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":50.52,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image","Depth"],"build":"202611032015107e","updatedAt":"2026-11-04T20:59:53.000Z"},{"model":"depth-anything/Depth-Anything-V2-Small-hf","hardware":"orin-nx-16gb","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":15.732336044311523,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true}]},{"title":"Accuracy","subtitle":"Hypersim","metrics":[{"description":"AbsRel","value":0.060633683929646645,"higherIsBetter":false,"digits":4,"headline":true},{"description":"δ1","value":96.89516391344806,"unit":"%","higherIsBetter":true,"digits":2}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":766,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":50.78,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image","Depth"],"build":"20261103201551b1","updatedAt":"2026-11-04T20:59:53.000Z"},{"model":"depth-anything/Depth-Anything-V2-Small-hf","hardware":"orin-nx-16gb","variant":"huggingface_fp16_tensorrt","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":15.732336044311523,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true}]},{"title":"Accuracy","subtitle":"Hypersim","metrics":[{"description":"AbsRel","value":0.060633683929646645,"higherIsBetter":false,"digits":4,"headline":true},{"description":"δ1","value":96.89516391344806,"unit":"%","higherIsBetter":true,"digits":2}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":766,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":50.78,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image","Depth"],"build":"20261103201551b1","updatedAt":"2026-11-04T20:59:53.000Z"},{"model":"depth-anything/Depth-Anything-V2-Small-hf","hardware":"orin-nano-super","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":19.340896606445312,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true}]},{"title":"Accuracy","subtitle":"Hypersim","metrics":[{"description":"AbsRel","value":0.060640767127166,"higherIsBetter":false,"digits":4,"headline":true},{"description":"δ1","value":96.89876530721261,"unit":"%","higherIsBetter":true,"digits":2}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":662.8,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":52.73,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image","Depth"],"build":"202611032015297f","updatedAt":"2026-11-04T20:59:53.000Z"},{"model":"depth-anything/Depth-Anything-V2-Small-hf","hardware":"orin-nano-super","variant":"huggingface_fp16_tensorrt","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":19.340896606445312,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true}]},{"title":"Accuracy","subtitle":"Hypersim","metrics":[{"description":"AbsRel","value":0.060640767127166,"higherIsBetter":false,"digits":4,"headline":true},{"description":"δ1","value":96.89876530721261,"unit":"%","higherIsBetter":true,"digits":2}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":662.8,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":52.73,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image","Depth"],"build":"202611032015297f","updatedAt":"2026-11-04T20:59:53.000Z"}]},{"name":"embedl/dinov3-vitb16","upstream":"facebook/dinov3-vitb16-pretrain-lvd1689m","title":"DINOv3 ViT-B/16","source":"embedl/dinov3-vitb16","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=embedl%2Fdinov3-vitb16&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"embedl/dinov3-vitb16","hardware":"thor","precision":"INT8","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":1.0703359842300415,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.2845439910888672,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Accuracy","subtitle":"Imagenette val","metrics":[{"description":"k-NN top-1","value":99.21019108280255,"unit":"%","digits":2,"baseline":99.7452229299363,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":562.8,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":623.8,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":85.56,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":165.26,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image"],"build":"202611021748c7d6","updatedAt":"2026-11-03T10:41:26.000Z"},{"model":"embedl/dinov3-vitb16","hardware":"thor","variant":"huggingface_fp16_tensorrt","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":1.2845439910888672,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true}]},{"title":"Accuracy","subtitle":"Imagenette val","metrics":[{"description":"k-NN top-1","value":99.7452229299363,"unit":"%","digits":2,"headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":623.8,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":165.26,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image"],"build":"202611021748c578","updatedAt":"2026-11-03T10:41:26.000Z"},{"model":"embedl/dinov3-vitb16","hardware":"thor","variant":"huggingface_int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":1.0703359842300415,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.2845439910888672,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Accuracy","subtitle":"Imagenette val","metrics":[{"description":"k-NN top-1","value":99.21019108280255,"unit":"%","digits":2,"baseline":99.7452229299363,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":562.8,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":623.8,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":85.56,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":165.26,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image"],"build":"202611021748c7d6","updatedAt":"2026-11-03T10:41:26.000Z"},{"model":"embedl/dinov3-vitb16","hardware":"orin","precision":"INT8","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":2.560447931289673,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.0234400033950806,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Accuracy","subtitle":"Imagenette val","metrics":[{"description":"k-NN top-1","value":99.3375796178344,"unit":"%","digits":2,"baseline":99.7452229299363,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":706.1,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":877.4,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":89.16,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":165.89,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image"],"build":"202611021307cc92","updatedAt":"2026-11-03T10:42:18.000Z"},{"model":"embedl/dinov3-vitb16","hardware":"orin","variant":"huggingface_fp16_tensorrt","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":3.0234400033950806,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true}]},{"title":"Accuracy","subtitle":"Imagenette val","metrics":[{"description":"k-NN top-1","value":99.7452229299363,"unit":"%","digits":2,"headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":877.4,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":165.89,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image"],"build":"20261102130709a5","updatedAt":"2026-11-03T10:42:18.000Z"},{"model":"embedl/dinov3-vitb16","hardware":"orin","variant":"huggingface_int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":2.560447931289673,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.0234400033950806,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Accuracy","subtitle":"Imagenette val","metrics":[{"description":"k-NN top-1","value":99.3375796178344,"unit":"%","digits":2,"baseline":99.7452229299363,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":706.1,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":877.4,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":89.16,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":165.89,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image"],"build":"202611021307cc92","updatedAt":"2026-11-03T10:42:18.000Z"},{"model":"embedl/dinov3-vitb16","hardware":"orin-nx-16gb","precision":"INT8","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":4.331087827682495,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.489344120025635,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Accuracy","subtitle":"Imagenette val","metrics":[{"description":"k-NN top-1","value":99.3375796178344,"unit":"%","digits":2,"baseline":99.7452229299363,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":537.7,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":853.4,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":89.2,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":167.88,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image"],"build":"2026110213269992","updatedAt":"2026-11-03T10:42:18.000Z"},{"model":"embedl/dinov3-vitb16","hardware":"orin-nx-16gb","variant":"huggingface_fp16_tensorrt","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":5.489344120025635,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true}]},{"title":"Accuracy","subtitle":"Imagenette val","metrics":[{"description":"k-NN top-1","value":99.7452229299363,"unit":"%","digits":2,"headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":853.4,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":167.88,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image"],"build":"202611021326a1b1","updatedAt":"2026-11-03T10:42:18.000Z"},{"model":"embedl/dinov3-vitb16","hardware":"orin-nx-16gb","variant":"huggingface_int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":4.331087827682495,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.489344120025635,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Accuracy","subtitle":"Imagenette val","metrics":[{"description":"k-NN top-1","value":99.3375796178344,"unit":"%","digits":2,"baseline":99.7452229299363,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":537.7,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":853.4,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":89.2,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":167.88,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image"],"build":"2026110213269992","updatedAt":"2026-11-03T10:42:18.000Z"},{"model":"embedl/dinov3-vitb16","hardware":"orin-nano-super","precision":"INT8","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":4.89356803894043,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":6.176640033721924,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Accuracy","subtitle":"Imagenette val","metrics":[{"description":"k-NN top-1","value":99.3375796178344,"unit":"%","digits":2,"baseline":99.77070063694268,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":673,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":828.1,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":89.42,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":165.85,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image"],"build":"20261102173617b5","updatedAt":"2026-11-03T10:41:08.000Z"},{"model":"embedl/dinov3-vitb16","hardware":"orin-nano-super","variant":"huggingface_fp16_tensorrt","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":6.176640033721924,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true}]},{"title":"Accuracy","subtitle":"Imagenette val","metrics":[{"description":"k-NN top-1","value":99.77070063694268,"unit":"%","digits":2,"headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":828.1,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":165.85,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image"],"build":"202611021736566d","updatedAt":"2026-11-03T10:41:08.000Z"},{"model":"embedl/dinov3-vitb16","hardware":"orin-nano-super","variant":"huggingface_int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":4.89356803894043,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":6.176640033721924,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Accuracy","subtitle":"Imagenette val","metrics":[{"description":"k-NN top-1","value":99.3375796178344,"unit":"%","digits":2,"baseline":99.77070063694268,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":673,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":828.1,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":89.42,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":165.85,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image"],"build":"20261102173617b5","updatedAt":"2026-11-03T10:41:08.000Z"}]},{"name":"timm/dla102.in1k","title":"DLA-102","source":"timm/dla102.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdla102.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/dla102.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.825471,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.912354,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1198,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.1,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.47,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b182801fcaca40b2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla102.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.67146,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.67871,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":372.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.1,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.47,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2e61a832d9722d3a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla102.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.24148,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.2511,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":234.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.1,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.47,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1fc0a38db91acc8c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla102.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.84866,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.936465,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.849976,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.94043,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1176.5,"unit":"qps","digits":1,"baseline":1066.25,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.1,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.47,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.89,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.37,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"7e61c33bf436488d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla102.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.61673,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.62012,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":618.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.1,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.47,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9d908f1b7d25a7a8","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/dla102x.in1k","title":"DLA-102-X","source":"timm/dla102x.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdla102x.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/dla102x.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.788331,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.89896,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.884949,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.976562,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1252.7,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1091.64,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.16,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.01,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4b0d9ed2120cd8d2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla102x.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.66451,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.67395,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":373.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.16,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.01,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d42682d5ae2dc93d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla102x.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.37199,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.38385,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":227.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.16,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.01,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"73aaa63380048179","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla102x.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.87428,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.875916,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1141.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.16,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.01,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.82,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.3,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"4cab6751dd58b277","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla102x.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.54753,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.55444,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":645.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.16,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.01,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"076c874c3b2070a5","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/dla102x2.in1k","title":"DLA-102x2","source":"timm/dla102x2.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdla102x2.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/dla102x2.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.996899,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.21474,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1.10236,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.31122,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":987.3,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":808.28,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.58,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.19,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"46626f28826b086d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla102x2.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.97793,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.98541,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":334,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.58,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.19,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ab4ce408762dc8f3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla102x2.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.20783,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.22162,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":191,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.58,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.19,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ed0d10f482c24a41","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla102x2.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.0813,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.08331,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":923.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.58,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.19,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.47,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"21ebd1a37704e44f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla102x2.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.69888,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.7063,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":588.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.58,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.19,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"239a44585731c8ac","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/dla34.in1k","title":"DLA-34","source":"timm/dla34.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdla34.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/dla34.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.30294,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.375085,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.358704,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.43634,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":3204.3,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":2544.81,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.3,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.93,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e5f7563799a2e3e5","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla34.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.00205,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.00677,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":991.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.3,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.93,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"fc97c8470d3ae1ff","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla34.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.67324,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.67981,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":594.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.3,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.93,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1b4f9f0da6a634bd","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla34.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.349872,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.350769,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2845.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":74.3,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.93,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.7,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.18,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"08b30cb206c97750","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla34.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.71688,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.718811,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1392.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.3,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.93,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"22dc5ad98916b486","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/dla46_c.in1k","title":"DLA-46-C","source":"timm/dla46_c.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdla46_c.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/dla46_c.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.293645,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.36908,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3355.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":63.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":85.63,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"04a526bbee773307","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla46_c.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.831174,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.836243,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1194.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":63.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":85.63,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"031d438f77b00323","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla46_c.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.36407,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.37134,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":728.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":63.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":85.63,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a4e3cdb4de912c2b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla46_c.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.266684,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.267395,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3728.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":63.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":85.63,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.54,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":15.8,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"0ac6d0fa610ae7dc","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla46_c.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.465683,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.46814,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2142.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":63.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":85.63,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"107c8b72f17ceba8","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/dla46x_c.in1k","title":"DLA-46-X-C","source":"timm/dla46x_c.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdla46x_c.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/dla46x_c.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.445199,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.516815,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2165.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":65.47,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":86.6,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"dd8746e4072dd477","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla46x_c.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.819924,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.823914,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1211.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":65.47,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":86.6,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c38ef1aec288e1d2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla46x_c.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.55565,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.56506,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":639,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":65.47,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":86.6,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"faa089544801e78b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla46x_c.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.318067,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.31897,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3129.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":65.47,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":86.6,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.54,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":15.9,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"04dac40bfe128281","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla46x_c.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.479997,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.483337,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2078.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":65.47,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":86.6,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"86ad6c5f2256a5b9","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/dla60.in1k","title":"DLA-60","source":"timm/dla60.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdla60.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/dla60.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.447006,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.582946,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.526489,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.653687,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":2206.4,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1665.54,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.44,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.96,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"996e22ce9ac74b11","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla60.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.56624,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.57477,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":634.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.44,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.96,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"47cc33bf4ea1782f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla60.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.63423,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.64221,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":377.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.44,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.96,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1605ad9af68546e6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla60.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.504163,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.505188,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1978,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.44,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.96,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.77,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.25,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"4d369f739a416c78","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla60.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.935571,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.939423,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1067.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.44,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.96,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a2a66754a5794240","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/dla60_res2net.in1k","title":"DLA-60 (res2net)","source":"timm/dla60_res2net.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdla60_res2net.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/dla60_res2net.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.773332,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.899453,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.873779,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.995361,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1285.5,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1095.56,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.02,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.08,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6afaeacb77f352a4","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla60_res2net.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.11283,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.11542,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":897.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.02,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.08,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.76,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.23,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"8a766719fa297416","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla60_res2net.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.92966,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.93628,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":517.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.02,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.08,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"894119c90df9b214","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/dla60_res2next.in1k","title":"DLA-60 (res2next)","source":"timm/dla60_res2next.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdla60_res2next.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/dla60_res2next.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.69836,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.849949,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.788086,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.941406,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1425.2,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1156.62,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.89,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8b7b9ea59c035714","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla60_res2next.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.43006,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.43982,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":290,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.89,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ff82b757e73dee5b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla60_res2next.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.57945,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.59326,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":178.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.89,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8061efa9a3af8879","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla60_res2next.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.09469,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.09628,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":912.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.89,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.71,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.19,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"4fdde7da51f88fc1","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla60_res2next.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.10916,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.11713,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":321.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.89,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a850b0186da9c656","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/dla60x.in1k","title":"DLA-60-X","source":"timm/dla60x.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdla60x.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/dla60x.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.460396,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.59966,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.526489,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.661743,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":2109.7,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1606.34,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.59,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.66,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"830fb8d83ebb47f6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla60x.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.54267,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.5517,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":644.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.59,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.66,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a2fb17e1df51c0a9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla60x.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.64201,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.65009,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":376.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.59,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.66,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0994b0fba6e73db0","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla60x.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.472316,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.473267,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2110.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.59,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.66,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.72,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.2,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"31bf66d4a869fce7","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla60x.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.846952,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.849915,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1179.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.59,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.66,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6032e04095db2b6d","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/dla60x_c.in1k","title":"DLA-60-X-C","source":"timm/dla60x_c.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdla60x_c.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/dla60x_c.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.500161,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.582703,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1938.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":67.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":88.07,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9f44d9e436741909","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla60x_c.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.949981,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.955627,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1045.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":67.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":88.07,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a3d38738ac2543f2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla60x_c.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.76556,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.77405,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":563.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":67.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":88.07,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"dad5c763f4bdefff","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla60x_c.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.412016,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.412964,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2417,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":67.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":88.07,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.54,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":19.5,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"77668927b0c25af9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dla60x_c.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.604583,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.607788,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1651.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":67.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":88.07,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d56a79e3cf33e4b9","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/dpn107.mx_in1k","title":"DPN-107","source":"timm/dpn107.mx_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdpn107.mx_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/dpn107.mx_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.53664,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.86567,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1.62946,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.96716,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":640.7,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":529.546,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.77,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"abcb4671b9f869a0","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dpn107.mx_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.88138,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.92041,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":169.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.77,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0c262880804c7952","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dpn107.mx_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":12.2814,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":12.3278,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":81,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.77,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4bffdca275d8b4cd","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dpn107.mx_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.74446,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.10436,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.74792,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.17499,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":572.8,"unit":"qps","digits":1,"baseline":474.877,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.77,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.5,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.98,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"ffce3c4ef752518a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dpn107.mx_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.66145,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.67035,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":375.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.77,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"92e199f8790ef11d","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/dpn131.mx_in1k","title":"DPN-131","source":"timm/dpn131.mx_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdpn131.mx_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/dpn131.mx_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.52105,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.80879,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1.63733,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.9101,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":646.5,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":543.729,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.62,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"77856913ebebf7ae","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dpn131.mx_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.18138,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.20215,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":160.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.62,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"22a1085a1a55865b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dpn131.mx_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":12.2858,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":12.3223,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":81,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.62,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"39ae02771ef58a57","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dpn131.mx_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.79769,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.07661,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.80316,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.11786,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":555.8,"unit":"qps","digits":1,"baseline":481.151,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.62,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.41,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.89,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"1b94add5fea88e3b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dpn131.mx_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.79209,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.80164,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":358,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.62,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"cc5f3e29dc13c5ad","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/dpn68.mx_in1k","title":"DPN-68","source":"timm/dpn68.mx_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdpn68.mx_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/dpn68.mx_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.475233,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.549255,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2057.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.74,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.7,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2787b09c071c32ac","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dpn68.mx_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.88273,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.88947,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":528.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.74,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.7,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ef78ffa12e82799c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dpn68.mx_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.43353,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.44507,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":289.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.74,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.7,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6657ed38750bc560","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dpn68.mx_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.550573,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.55127,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1811.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":75.74,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.7,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.66,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.14,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"8c18dc8c23597e8c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dpn68.mx_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.97783,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.980225,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1021.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.74,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.7,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"62c4416a5c142850","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/dpn68b.mx_in1k","title":"DPN-68-B","source":"timm/dpn68b.mx_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdpn68b.mx_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/dpn68b.mx_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.588307,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.661438,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1686,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.28,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.16,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e84e7f430a58d48f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dpn68b.mx_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.28229,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.29132,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":435.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.28,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.16,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"646553a162d1f0ed","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dpn68b.mx_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.95971,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.97119,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":251.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.28,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.16,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f24d4d56f85b6d54","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dpn68b.mx_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.630576,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.631409,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1582.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.28,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.16,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.66,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.14,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"f2108e11ed67c8e0","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dpn68b.mx_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.0899,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.11572,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":916.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.28,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.16,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f4eb29bede4285a1","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/dpn92.mx_in1k","title":"DPN-92","source":"timm/dpn92.mx_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdpn92.mx_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/dpn92.mx_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.18317,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.41028,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":237.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.66,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.63,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5548dc67a76ae860","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dpn92.mx_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.70121,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.73962,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":148.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.66,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.63,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e57b31bbb1213c10","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dpn92.mx_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":13.9521,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":13.9709,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":71.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.66,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.63,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"de99477f5c8f460b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dpn92.mx_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.06648,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.14978,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":325.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.66,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.63,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.05,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.53,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"018c56a4886fad2d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dpn92.mx_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.60943,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.64746,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":277,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.66,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.63,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"25a17c2e0db93a39","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/dpn98.mx_in1k","title":"DPN-98","source":"timm/dpn98.mx_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdpn98.mx_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/dpn98.mx_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.17833,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.40387,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1.28113,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.51672,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":829.2,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":697.975,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.43,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.47,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"04fe0494f5fe8247","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dpn98.mx_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.59007,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.60596,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":216.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.43,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.47,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8b90b0c2743ec07b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dpn98.mx_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.22079,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":9.23877,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":107.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.43,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.47,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"44390a3190b74eb6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dpn98.mx_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.35157,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.56471,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.35507,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.5683,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":739.1,"unit":"qps","digits":1,"baseline":638.541,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.43,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.47,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.21,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.69,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"c46f510744324513","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/dpn98.mx_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.13538,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.15045,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":468.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.43,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.47,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ed3c1c6c0e70016f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/ecaresnet101d.miil_in1k","title":"ECA-ResNet-101-D","source":"timm/ecaresnet101d.miil_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fecaresnet101d.miil_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/ecaresnet101d.miil_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.64916,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.66504,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":605.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.02,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.79,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.02,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.5,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"f9cda3d9dd2220cf","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/ecaresnet26t.ra2_in1k","title":"ECA-ResNet-26-T","source":"timm/ecaresnet26t.ra2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fecaresnet26t.ra2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/ecaresnet26t.ra2_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.380881,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.521598,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.450378,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.582825,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":2540.8,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1831.7,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.45,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.36,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3b49f212cc479dcc","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/ecaresnet26t.ra2_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.47647,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.48352,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":673.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.45,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.36,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2d2245f67e27406f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/ecaresnet26t.ra2_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.0258,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.04541,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":328.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.45,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.36,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5df06cd25937a512","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/ecaresnet26t.ra2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.442899,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.444519,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2250.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.45,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.36,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.7,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.18,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"1fd3694605de653f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/ecaresnet26t.ra2_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.763063,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.765717,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1308.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.45,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.36,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"78431028d3251848","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/eca_resnet33ts.ra2_in1k","title":"ECA-ResNet-33-TS","source":"timm/eca_resnet33ts.ra2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Feca_resnet33ts.ra2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/eca_resnet33ts.ra2_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.60625,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.668823,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1584.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.37,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.62,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1b6a8c43729318b3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eca_resnet33ts.ra2_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.82297,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.83466,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":545.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.37,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.62,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"40801e0c3018acb8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eca_resnet33ts.ra2_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.34342,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.35657,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":297.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.37,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.62,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"25374726e1c50648","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eca_resnet33ts.ra2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.38388,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.38965,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":721.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.37,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.62,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.81,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.28,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"834610caf8ab2e78","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eca_resnet33ts.ra2_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.61739,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.62469,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":617.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.37,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.62,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6e21e371cdc8a788","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/ecaresnet50d.miil_in1k","title":"ECA-ResNet-50-D","source":"timm/ecaresnet50d.miil_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fecaresnet50d.miil_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/ecaresnet50d.miil_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.768912,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.771301,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1297.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.34,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.73,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.81,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.29,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"ade3f10f43f781c7","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/ecaresnet50t.a1_in1k","title":"ECA-ResNet-50-T","source":"timm/ecaresnet50t.a1_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fecaresnet50t.a1_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/ecaresnet50t.a1_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.72368,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.827768,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.809265,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.908569,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1350.6,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1181,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.54,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.85,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1e2f72fc58d5d8fe","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/ecaresnet50t.a1_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.42055,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.42749,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":410.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.54,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.85,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"fab598347a626990","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/ecaresnet50t.a1_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.78228,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.7948,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":208,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.54,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.85,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"135d490a7f4600ca","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/ecaresnet50t.a1_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.723994,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.725708,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1378.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.54,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.85,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.81,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.29,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"a9379e55415f7e0f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/ecaresnet50t.a1_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.20617,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.20886,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":828.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.54,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.85,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b2ce0e48643ce003","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/eca_resnext26ts.ch_in1k","title":"ECA-ResNeXt-26-TS","source":"timm/eca_resnext26ts.ch_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Feca_resnext26ts.ch_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/eca_resnext26ts.ch_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.445039,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.502869,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2130.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.25,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.48,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d9b3231d90ee6356","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eca_resnext26ts.ch_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.21501,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.22168,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":818,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.25,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.48,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"117588a6a658142d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eca_resnext26ts.ch_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.29623,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.30377,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":433,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.25,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.48,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0387585fd6da986d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eca_resnext26ts.ch_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.642772,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.646118,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1552.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.25,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.48,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.66,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.14,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e580446769af348b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eca_resnext26ts.ch_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.847711,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.85614,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1178.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.25,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.48,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f7c5ec7865d0970a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/edgenext_base.in21k_ft_in1k","title":"EdgeNeXt Base","source":"timm/edgenext_base.in21k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fedgenext_base.in21k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/edgenext_base.in21k_ft_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.83134,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.978393,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.926636,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.07483,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1174.8,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":991.801,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.72,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"353bd0f4819f5f8a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/edgenext_base.in21k_ft_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.0584,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.0683,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":483.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.72,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c7e01132ebd4f679","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/edgenext_base.in21k_ft_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.93049,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.94519,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":253.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.72,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"44f4277bde23f853","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/edgenext_base.in21k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.748982,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.753662,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1332.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.72,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.74,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.21,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"3a8c51e8cc9e715c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/edgenext_base.in21k_ft_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.15961,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.16736,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":861.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.72,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"77514980f42611c4","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/edgenext_small.usi_in1k","title":"EdgeNeXt Small","source":"timm/edgenext_small.usi_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fedgenext_small.usi_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/edgenext_small.usi_in1k","hardware":"l4","variant":"usi_in1k-timm_int8_embedl_tensorrt_l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.550443,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.551453,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1811.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNetValFullDataset","metrics":[{"description":"Top-1 (ImageNetValFullDataset)","value":80.59,"unit":"%","digits":2,"headline":true},{"description":"Top-5 (ImageNetValFullDataset)","value":95.23,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"054163824bc1cf1e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/edgenext_small_rw.sw_in1k","title":"EdgeNeXt Small (rw)","source":"timm/edgenext_small_rw.sw_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fedgenext_small_rw.sw_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/edgenext_small_rw.sw_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.27605,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.38391,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":302.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.38,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"906e13f3f1858688","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/edgenext_small_rw.sw_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.04758,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":9.06506,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":110.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.38,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"03ac0fb356d237d6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/edgenext_small_rw.sw_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":12.4995,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":12.5333,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":79.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.38,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"64d4773ab5d468af","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/edgenext_small_rw.sw_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.59553,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.59656,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":278,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.38,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.61,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":92.5,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"6bdb5a05a5f0b1be","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/edgenext_small_rw.sw_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.27073,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.27539,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":159.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.38,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"93ef2bb3b5ca455c","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/edgenext_x_small.in1k","title":"EdgeNeXt XSmall","source":"timm/edgenext_x_small.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fedgenext_x_small.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/edgenext_x_small.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.614963,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.698975,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1612.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.61,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.92,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"88c9568265a4c536","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/edgenext_x_small.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.35579,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.36273,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":733.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.61,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.92,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c78eb60f4ee1dd12","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/edgenext_x_small.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.14611,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.15332,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":463.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.61,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.92,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"885e48ae20a80f5e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/edgenext_x_small.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.472109,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.472778,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2110.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":74.61,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.92,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.55,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":29.3,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"41593d3664b3c3ef","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/edgenext_x_small.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.691929,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.694336,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1443,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.61,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.92,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1fa81297ab9105c0","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/edgenext_xx_small.in1k","title":"EdgeNeXt XXSmall","source":"timm/edgenext_xx_small.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fedgenext_xx_small.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/edgenext_xx_small.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.476166,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.562744,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2076.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":70.54,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.55,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b3e3192aaf7fda91","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/edgenext_xx_small.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.07651,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.08362,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":923,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":70.54,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.55,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"38d0ae05d6306a8c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/edgenext_xx_small.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.64801,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.65448,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":603.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":70.54,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.55,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"606d9abbf23ee97b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/edgenext_xx_small.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.372021,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.37262,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2675.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":70.54,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.55,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.54,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":17.2,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"da567f3976e7e81c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/edgenext_xx_small.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.532868,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.534546,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1872.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":70.54,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.55,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1870c0e7cb0ded84","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/efficientformer_l1.snap_dist_in1k","title":"EfficientFormer-L1","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fefficientformer_l1.snap_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/efficientformer_l1.snap_dist_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":485.168,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":485.962,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":2055.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":19.72,"unit":"%","headline":true},{"description":"Top-5","value":36.36,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/efficientformer_l7.snap_dist_in1k","title":"EfficientFormer-L7","source":"timm/efficientformer_l7.snap_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fefficientformer_l7.snap_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/efficientformer_l7.snap_dist_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.73247,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.98923,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1.8476,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.12311,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":569.5,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":496.93,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.7,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.14,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c95cbb8713992233","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientformer_l7.snap_dist_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.34564,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.36005,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":228.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.7,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.14,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9005f20e6842570a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientformer_l7.snap_dist_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.57669,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":8.59784,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":116,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.7,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.14,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5ebc00961d66086d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientformer_l7.snap_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.7203,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.72821,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":580.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.7,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.14,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.47,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.95,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"ba940b0871b69409","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientformer_l7.snap_dist_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.58821,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.60449,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":386.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.7,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.14,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f1648287a9931252","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/efficientformerv2_s0.snap_dist_in1k","title":"EfficientFormerV2-S0","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fefficientformerv2_s0.snap_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/efficientformerv2_s0.snap_dist_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":662.292,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":664.062,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":1506.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.14,"unit":"%","headline":true},{"description":"Top-5","value":0.82,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/efficientformerv2_s1.snap_dist_in1k","title":"EfficientFormerV2-S1","source":"timm/efficientformerv2_s1.snap_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fefficientformerv2_s1.snap_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/efficientformerv2_s1.snap_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency on NVIDIA L4","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.79134,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.793091,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1261.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.35,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.54,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.6,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.08,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"b951dba5fdb4b0f5","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/efficientformerv2_s2.snap_dist_in1k","title":"EfficientFormerV2-S2","source":"timm/efficientformerv2_s2.snap_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fefficientformerv2_s2.snap_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/efficientformerv2_s2.snap_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.21974,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.22217,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":818.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.89,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.7,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.69,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.17,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"0187c8997acb03c8","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/efficientnet_b0","title":"EfficientNet-B0","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fefficientnet_b0&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/efficientnet_b0","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":822.449,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":825.623,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":1213.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.67,"unit":"%","headline":true},{"description":"Top-5","value":2.05,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/efficientnet_b0","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":822.449,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":825.623,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":1213.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.67,"unit":"%","headline":true},{"description":"Top-5","value":2.05,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tf_efficientnet_b0.aa_in1k","title":"EfficientNet-B0 (TF)","source":"timm/tf_efficientnet_b0.aa_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftf_efficientnet_b0.aa_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tf_efficientnet_b0.aa_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.749119,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.75238,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1332.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.95,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.07,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"4b3d53186543a9c8","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/efficientnet_b0.ra4_e3600_r224_in1k","title":"EfficientNet-B0 (timm)","source":"timm/efficientnet_b0.ra4_e3600_r224_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fefficientnet_b0.ra4_e3600_r224_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/efficientnet_b0.ra4_e3600_r224_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.615131,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.618164,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1622.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.19,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.17,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.07,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"60ba3a7b27fd7e43","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/efficientnet_b1","title":"EfficientNet-B1","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fefficientnet_b1&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/efficientnet_b1","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.207,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.209,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":827.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.09,"unit":"%","headline":true},{"description":"Top-5","value":0.49,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/efficientnet_b1","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.207,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.209,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":827.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.09,"unit":"%","headline":true},{"description":"Top-5","value":0.49,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/efficientnet_b1_pruned.in1k","title":"EfficientNet-B1 (pruned)","source":"timm/efficientnet_b1_pruned.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fefficientnet_b1_pruned.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/efficientnet_b1_pruned.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.97156,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.974182,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1027.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.45,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.51,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.6,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.08,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"cad9bcdf776a7300","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tf_efficientnet_b1.aa_in1k","title":"EfficientNet-B1 (TF)","source":"timm/tf_efficientnet_b1.aa_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftf_efficientnet_b1.aa_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tf_efficientnet_b1.aa_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.12127,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.12585,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":890.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.61,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.02,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.62,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.1,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"36273020d58de2cb","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/efficientnet_b1.ft_in1k","title":"EfficientNet-B1 (timm)","source":"timm/efficientnet_b1.ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fefficientnet_b1.ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/efficientnet_b1.ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.04576,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.04828,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":955,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.46,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.51,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.62,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.1,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"b00c3cbcb23a4465","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/efficientnet_b2","title":"EfficientNet-B2","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fefficientnet_b2&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/efficientnet_b2","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.229,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.231,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":812.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":45.43,"unit":"%","headline":true},{"description":"Top-5","value":68.87,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/efficientnet_b2","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.229,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.231,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":812.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":45.43,"unit":"%","headline":true},{"description":"Top-5","value":68.87,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tf_efficientnet_b2.aa_in1k","title":"EfficientNet-B2 (TF)","source":"timm/tf_efficientnet_b2.aa_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftf_efficientnet_b2.aa_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tf_efficientnet_b2.aa_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.27396,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.27942,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":784,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.71,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.73,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.64,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.11,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"4012a2757e3790ad","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/efficientnet_b2.ra_in1k","title":"EfficientNet-B2 (timm)","source":"timm/efficientnet_b2.ra_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fefficientnet_b2.ra_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/efficientnet_b2.ra_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.16414,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.16754,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":858,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.54,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.24,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.64,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.11,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"84461c13ac8fc6d2","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/efficientnet_b3","title":"EfficientNet-B3","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fefficientnet_b3&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/efficientnet_b3","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.437,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.44,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":695,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":16.99,"unit":"%","headline":true},{"description":"Top-5","value":31.47,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/efficientnet_b3","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.437,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.44,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":695,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":16.99,"unit":"%","headline":true},{"description":"Top-5","value":31.47,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/efficientnet_b3_pruned.in1k","title":"EfficientNet-B3 (pruned)","source":"timm/efficientnet_b3_pruned.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fefficientnet_b3_pruned.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/efficientnet_b3_pruned.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.36786,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.37177,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":730.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.11,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.97,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.65,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.12,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"0f4bfaf8ad1e6b94","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/efficientnet_b3.ra2_in1k","title":"EfficientNet-B3 (timm)","source":"timm/efficientnet_b3.ra2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fefficientnet_b3.ra2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/efficientnet_b3.ra2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.70619,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.7168,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":585.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.32,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.68,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.15,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e5f18cb090a0fffa","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/efficientnet_b4","title":"EfficientNet-B4","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fefficientnet_b4&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/efficientnet_b4","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.773,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.781,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":563.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.11,"unit":"%","headline":true},{"description":"Top-5","value":0.52,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/efficientnet_b4","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.773,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.781,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":563.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.11,"unit":"%","headline":true},{"description":"Top-5","value":0.52,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/efficientnet_b5","title":"EfficientNet-B5","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fefficientnet_b5&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/efficientnet_b5","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":2.597,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":2.69,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":382.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":53.53,"unit":"%","headline":true},{"description":"Top-5","value":75.68,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/efficientnet_b5","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":2.597,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":2.69,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":382.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":53.53,"unit":"%","headline":true},{"description":"Top-5","value":75.68,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/efficientnet_b6","title":"EfficientNet-B6","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fefficientnet_b6&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/efficientnet_b6","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":3.168,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":3.28,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":313.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":64.99,"unit":"%","headline":true},{"description":"Top-5","value":85.78,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/efficientnet_b6","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":3.168,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":3.28,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":313.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":64.99,"unit":"%","headline":true},{"description":"Top-5","value":85.78,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/efficientnet_b7","title":"EfficientNet-B7","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fefficientnet_b7&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/efficientnet_b7","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":4.048,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":4.203,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":244.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":67.32,"unit":"%","headline":true},{"description":"Top-5","value":87.51,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/efficientnet_b7","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":4.048,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":4.203,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":244.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":67.32,"unit":"%","headline":true},{"description":"Top-5","value":87.51,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tf_efficientnet_cc_b0_4e.in1k","title":"EfficientNet-CC (b0, 4e, TF)","source":"timm/tf_efficientnet_cc_b0_4e.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftf_efficientnet_cc_b0_4e.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tf_efficientnet_cc_b0_4e.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.2003,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.20392,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":832.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.77,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.98,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.7,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.18,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"39cadd88f30178cc","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/efficientnet_el.ra_in1k","title":"EfficientNet-EL","source":"timm/efficientnet_el.ra_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fefficientnet_el.ra_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/efficientnet_el.ra_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.657701,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.848288,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.717102,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.917786,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1479.6,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1143.88,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.22,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"46345156b39142e1","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientnet_el.ra_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.23005,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.90386,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1.23645,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.91162,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":808,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":522.243,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.22,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d35d180548e6268e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientnet_el.ra_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.60067,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.62366,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":382.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.22,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9fc071dfded3ca51","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientnet_el.ra_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.404748,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.698297,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.407227,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.703796,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":2459.3,"unit":"qps","digits":1,"baseline":1428.66,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.22,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.64,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.12,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"3291bac3d7e47a3c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientnet_el.ra_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.602623,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.60437,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1656.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.22,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1add227f8610d5bc","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/efficientnet_el_pruned.in1k","title":"EfficientNet-EL (pruned)","source":"timm/efficientnet_el_pruned.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fefficientnet_el_pruned.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/efficientnet_el_pruned.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.657215,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.855335,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.718384,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.937195,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1482.9,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1139.13,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.19,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.69,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b7aae8d9720e61b7","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientnet_el_pruned.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.23273,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.91323,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1.23804,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.91943,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":806.2,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":519.68,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.19,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.69,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"daa24a5ec6c256a5","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientnet_el_pruned.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.61336,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.62085,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":380.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.19,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.69,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a6c22630598f4936","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientnet_el_pruned.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.37404,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.67709,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.375793,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.682251,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":2663.4,"unit":"qps","digits":1,"baseline":1458.58,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.19,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.69,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":65.2,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"af02ed8293472851","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientnet_el_pruned.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.599469,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.601135,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1665.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.19,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.69,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7c367a5590e834f4","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tf_efficientnet_el.in1k","title":"EfficientNet-EL (TF)","source":"timm/tf_efficientnet_el.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftf_efficientnet_el.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tf_efficientnet_el.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.402617,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.695429,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.404114,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.699646,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":2473.4,"unit":"qps","digits":1,"baseline":1435.18,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.46,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.74,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.64,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.12,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"fa1268a438d8478e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/efficientnet_em.ra2_in1k","title":"EfficientNet-EM","source":"timm/efficientnet_em.ra2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fefficientnet_em.ra2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/efficientnet_em.ra2_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.384402,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.473886,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.457275,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.552307,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":2529.9,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":2042.25,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.65,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.48,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d565f205d495fe7f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientnet_em.ra2_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.756391,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.10827,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.761169,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.11429,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1312.8,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":896.645,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.65,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.48,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6412d86debf8b302","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientnet_em.ra2_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.382,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.38934,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":719.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.65,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.48,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9e8aeb77dfb4202e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientnet_em.ra2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.240332,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.361852,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.241272,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.363464,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":4133.7,"unit":"qps","digits":1,"baseline":2752.92,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.65,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.48,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.6,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":81.5,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"9130b950ae9bbe51","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientnet_em.ra2_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.435626,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.436646,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2290,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.65,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.48,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"fda6f5e0fa23600a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tf_efficientnet_em.in1k","title":"EfficientNet-EM (TF)","source":"timm/tf_efficientnet_em.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftf_efficientnet_em.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tf_efficientnet_em.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.236942,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.37914,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.237854,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.382385,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":4192.4,"unit":"qps","digits":1,"baseline":2578.54,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.79,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.6,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":81.2,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"66c6ae8d1f4f1fef","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/efficientnet_es.ra_in1k","title":"EfficientNet-ES","source":"timm/efficientnet_es.ra_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fefficientnet_es.ra_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/efficientnet_es.ra_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.283748,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.341355,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.336609,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.412811,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":3422.9,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":2811.52,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.73,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.34,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"85b6a11ff290097c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientnet_es.ra_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.565747,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.823414,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.569092,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.828369,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1754.2,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1206.2,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.73,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.34,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1555b27ad27fd0f6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientnet_es.ra_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.991096,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.996887,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1002.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.73,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.34,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4262b37175ca65b1","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientnet_es.ra_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.176802,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.262526,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.17749,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.263855,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":5614.4,"unit":"qps","digits":1,"baseline":3789.72,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.73,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.34,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.58,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":64.1,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"180190278f044d54","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientnet_es.ra_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.324779,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.327087,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3068.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.73,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.34,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"44fd42ed85a2c7c6","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/efficientnet_es_pruned.in1k","title":"EfficientNet-ES (pruned)","source":"timm/efficientnet_es_pruned.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fefficientnet_es_pruned.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/efficientnet_es_pruned.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.279141,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.339313,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.336365,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.41272,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":3485.3,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":2825.75,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":73.65,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.92,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"afa8529f886782a1","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientnet_es_pruned.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.560585,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.823352,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.565063,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.829956,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1770.2,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1206.27,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":73.65,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.92,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"66766ee5dabb83e2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientnet_es_pruned.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.988022,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.991699,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1005.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":73.65,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.92,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"949a70ed5de3af11","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientnet_es_pruned.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.177485,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.25484,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.178406,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.255859,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":5584.3,"unit":"qps","digits":1,"baseline":3897.31,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":73.65,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.92,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.55,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":24.9,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"68d5b7d95617d0c8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientnet_es_pruned.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.324786,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.326599,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3069.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":73.65,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.92,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d11fd008a58600d1","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tf_efficientnet_es.in1k","title":"EfficientNet-ES (TF)","source":"timm/tf_efficientnet_es.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftf_efficientnet_es.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tf_efficientnet_es.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.178712,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.261451,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.179688,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.262756,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":5553.7,"unit":"qps","digits":1,"baseline":3804.77,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.02,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.78,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.58,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":64.1,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"ec27d00b5468e126","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/efficientnet_lite0.ra_in1k","title":"EfficientNet-Lite0","source":"timm/efficientnet_lite0.ra_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fefficientnet_lite0.ra_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/efficientnet_lite0.ra_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.279292,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.339478,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3463.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.17,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.41,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"bc48d4c07383bdd9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientnet_lite0.ra_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.499176,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.620301,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.502686,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.625122,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1987.4,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1600.03,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.17,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.41,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b0a7965d7fa772da","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientnet_lite0.ra_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.798359,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.803711,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1243.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.17,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.41,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"18aae85cf8c76d2f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientnet_lite0.ra_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.153898,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.180652,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.154724,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.182526,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":6436.3,"unit":"qps","digits":1,"baseline":5493.18,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":75.17,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.41,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.58,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":54.9,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"c5c9e29f1cdceedd","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientnet_lite0.ra_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.2714,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.272583,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3670.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.17,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.41,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9230fcd4da6aae4a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tf_efficientnet_lite0.in1k","title":"EfficientNet-Lite0 (TF)","source":"timm/tf_efficientnet_lite0.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftf_efficientnet_lite0.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tf_efficientnet_lite0.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.155625,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.180368,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.156067,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.182617,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":6361,"unit":"qps","digits":1,"baseline":5502.49,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":74.9,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.12,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.57,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":54.6,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"c49dc5964445bcc4","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tf_efficientnet_lite1.in1k","title":"EfficientNet-Lite1 (TF)","source":"timm/tf_efficientnet_lite1.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftf_efficientnet_lite1.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tf_efficientnet_lite1.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.20103,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.244625,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.201843,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.246216,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":4934,"unit":"qps","digits":1,"baseline":4065.43,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.04,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.58,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":63.6,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"2696448373bd9f66","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tf_efficientnet_lite2.in1k","title":"EfficientNet-Lite2 (TF)","source":"timm/tf_efficientnet_lite2.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftf_efficientnet_lite2.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tf_efficientnet_lite2.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.214473,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.288623,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.215271,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.290588,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":4630.8,"unit":"qps","digits":1,"baseline":3445.14,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.59,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":71.4,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"2610a19dd1420bb2","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tf_efficientnet_lite3.in1k","title":"EfficientNet-Lite3 (TF)","source":"timm/tf_efficientnet_lite3.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftf_efficientnet_lite3.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tf_efficientnet_lite3.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.262016,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.415005,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.262756,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.41684,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":3796.4,"unit":"qps","digits":1,"baseline":2399.43,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.65,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.79,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.61,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":95.7,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"cddffe6a1a7d6ead","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tf_efficientnet_lite4.in1k","title":"EfficientNet-Lite4 (TF)","source":"timm/tf_efficientnet_lite4.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftf_efficientnet_lite4.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tf_efficientnet_lite4.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.411421,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.728554,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.413208,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.732544,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":2421.2,"unit":"qps","digits":1,"baseline":1369.62,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.59,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.67,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.15,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"2bf89778415fa175","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tf_efficientnetv2_b0.in1k","title":"EfficientNetV2-B0 (TF)","source":"timm/tf_efficientnetv2_b0.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftf_efficientnetv2_b0.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tf_efficientnetv2_b0.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.716841,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.719177,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1392.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.99,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.61,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.09,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"c3cf60ee25e81a1b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tf_efficientnetv2_b1.in1k","title":"EfficientNetV2-B1 (TF)","source":"timm/tf_efficientnetv2_b1.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftf_efficientnetv2_b1.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tf_efficientnetv2_b1.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.913213,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.914978,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1093.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.65,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.55,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.62,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.1,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"fa91816b660bb9aa","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/efficientnet_v2_l","title":"EfficientNetV2-L","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fefficientnet_v2_l&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/efficientnet_v2_l","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":4.542,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":4.594,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":220.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.1,"unit":"%","headline":true},{"description":"Top-5","value":0.52,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/efficientnet_v2_l","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":4.542,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":4.594,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":220.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.1,"unit":"%","headline":true},{"description":"Top-5","value":0.52,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/efficientnet_v2_m","title":"EfficientNetV2-M","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fefficientnet_v2_m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/efficientnet_v2_m","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":2.929,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":2.987,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":340.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.09,"unit":"%","headline":true},{"description":"Top-5","value":0.56,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/efficientnet_v2_m","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":2.929,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":2.987,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":340.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.09,"unit":"%","headline":true},{"description":"Top-5","value":0.56,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/efficientnet_v2_s","title":"EfficientNetV2-S","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fefficientnet_v2_s&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/efficientnet_v2_s","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.734,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.736,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":576.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.09,"unit":"%","headline":true},{"description":"Top-5","value":0.55,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/efficientnet_v2_s","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.734,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.736,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":576.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.09,"unit":"%","headline":true},{"description":"Top-5","value":0.55,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"embedl/efficienttam-s-512","upstream":"yunyangx/efficient-track-anything","title":"EfficientTAM-S (512)","source":"embedl/efficienttam-s-512","imageUrl":"https://storage.googleapis.com/embedl-model-media/sha256/59a1fa271c406451924bdb4aa6b627868228697ebb603eb7374bd54448fb109c.webp","builds":[{"model":"embedl/efficienttam-s-512","hardware":"thor","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":6.37,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true}]},{"title":"Accuracy","subtitle":"DAVIS 2017 val · 8 sequences, 544 frames","metrics":[{"description":"J&F","value":85.16042341428034,"higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":561.7,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":91.94,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Video","Segmentation"],"build":"2026110423149752","updatedAt":"2026-11-10T15:54:34.000Z"},{"model":"embedl/efficienttam-s-512","hardware":"thor","variant":"huggingface_fp16_tensorrt","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":6.37,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true}]},{"title":"Accuracy","subtitle":"DAVIS 2017 val · 8 sequences, 544 frames","metrics":[{"description":"J&F","value":85.16042341428034,"higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":561.7,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":91.94,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Video","Segmentation"],"build":"2026110423149752","updatedAt":"2026-11-10T15:54:34.000Z"},{"model":"embedl/efficienttam-s-512","hardware":"orin","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":17.257,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true}]},{"title":"Accuracy","subtitle":"DAVIS 2017 val · 8 sequences, 544 frames","metrics":[{"description":"J&F","value":85.1743070426937,"higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":852.5,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":90.28,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Video","Segmentation"],"build":"202611042314a091","updatedAt":"2026-11-10T15:54:34.000Z"},{"model":"embedl/efficienttam-s-512","hardware":"orin","variant":"huggingface_fp16_tensorrt","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":17.257,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true}]},{"title":"Accuracy","subtitle":"DAVIS 2017 val · 8 sequences, 544 frames","metrics":[{"description":"J&F","value":85.1743070426937,"higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":852.5,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":90.28,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Video","Segmentation"],"build":"202611042314a091","updatedAt":"2026-11-10T15:54:34.000Z"},{"model":"embedl/efficienttam-s-512","hardware":"orin-nano-super","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":34.257999999999996,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true}]},{"title":"Accuracy","subtitle":"DAVIS 2017 val · 8 sequences, 544 frames","metrics":[{"description":"J&F","value":85.18597071043135,"higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":846.8,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":89.69,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Video","Segmentation"],"build":"202611042314bc4b","updatedAt":"2026-11-10T15:54:34.000Z"},{"model":"embedl/efficienttam-s-512","hardware":"orin-nano-super","variant":"huggingface_fp16_tensorrt","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":34.257999999999996,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true}]},{"title":"Accuracy","subtitle":"DAVIS 2017 val · 8 sequences, 544 frames","metrics":[{"description":"J&F","value":85.18597071043135,"higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":846.8,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":89.69,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Video","Segmentation"],"build":"202611042314bc4b","updatedAt":"2026-11-10T15:54:34.000Z"},{"model":"embedl/efficienttam-s-512","hardware":"l4","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":6.45,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true}]},{"title":"Accuracy","subtitle":"DAVIS 2017 val · 8 sequences, 544 frames","metrics":[{"description":"J&F","value":85.14638870371792,"higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":583.2,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":95.03,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Video","Segmentation"],"build":"202611042314a965","updatedAt":"2026-11-10T15:54:34.000Z"},{"model":"embedl/efficienttam-s-512","hardware":"l4","variant":"huggingface_fp16_tensorrt","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":6.45,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true}]},{"title":"Accuracy","subtitle":"DAVIS 2017 val · 8 sequences, 544 frames","metrics":[{"description":"J&F","value":85.14638870371792,"higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":583.2,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":95.03,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Video","Segmentation"],"build":"202611042314a965","updatedAt":"2026-11-10T15:54:34.000Z"}]},{"name":"embedl/efficienttam-ti-512","title":"EfficientTAM-Ti (512)","source":"embedl/efficienttam-ti-512","imageUrl":"https://storage.googleapis.com/embedl-model-media/sha256/59a1fa271c406451924bdb4aa6b627868228697ebb603eb7374bd54448fb109c.webp","builds":[{"model":"embedl/efficienttam-ti-512","hardware":"thor","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":4.391,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true}]},{"title":"Accuracy","subtitle":"DAVIS 2017 val · 8 sequences, 544 frames","metrics":[{"description":"J&F","value":81.90150390571577,"higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":561.7,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":56.68,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Video","Segmentation"],"build":"20261104231424bf","updatedAt":"2026-11-10T15:54:34.000Z"},{"model":"embedl/efficienttam-ti-512","hardware":"thor","variant":"huggingface_fp16_tensorrt","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":4.391,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true}]},{"title":"Accuracy","subtitle":"DAVIS 2017 val · 8 sequences, 544 frames","metrics":[{"description":"J&F","value":81.90150390571577,"higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":561.7,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":56.68,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Video","Segmentation"],"build":"20261104231424bf","updatedAt":"2026-11-10T15:54:34.000Z"},{"model":"embedl/efficienttam-ti-512","hardware":"orin","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":12.232,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true}]},{"title":"Accuracy","subtitle":"DAVIS 2017 val · 8 sequences, 544 frames","metrics":[{"description":"J&F","value":81.90550044279345,"higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":827.7,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":55.18,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Video","Segmentation"],"build":"202611042314680f","updatedAt":"2026-11-10T15:54:34.000Z"},{"model":"embedl/efficienttam-ti-512","hardware":"orin","variant":"huggingface_fp16_tensorrt","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":12.232,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true}]},{"title":"Accuracy","subtitle":"DAVIS 2017 val · 8 sequences, 544 frames","metrics":[{"description":"J&F","value":81.90550044279345,"higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":827.7,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":55.18,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Video","Segmentation"],"build":"202611042314680f","updatedAt":"2026-11-10T15:54:34.000Z"},{"model":"embedl/efficienttam-ti-512","hardware":"orin-nano-super","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":23.174,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true}]},{"title":"Accuracy","subtitle":"DAVIS 2017 val · 8 sequences, 544 frames","metrics":[{"description":"J&F","value":81.89910494443559,"higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":737.9,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":55.15,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Video","Segmentation"],"build":"202611042314a561","updatedAt":"2026-11-10T15:54:34.000Z"},{"model":"embedl/efficienttam-ti-512","hardware":"orin-nano-super","variant":"huggingface_fp16_tensorrt","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":23.174,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true}]},{"title":"Accuracy","subtitle":"DAVIS 2017 val · 8 sequences, 544 frames","metrics":[{"description":"J&F","value":81.89910494443559,"higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":737.9,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":55.15,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Video","Segmentation"],"build":"202611042314a561","updatedAt":"2026-11-10T15:54:34.000Z"},{"model":"embedl/efficienttam-ti-512","hardware":"l4","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":4.835999999999999,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true}]},{"title":"Accuracy","subtitle":"DAVIS 2017 val · 8 sequences, 544 frames","metrics":[{"description":"J&F","value":81.88095696514159,"higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":583.1,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":59.730000000000004,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Video","Segmentation"],"build":"202611042314701a","updatedAt":"2026-11-10T15:54:34.000Z"},{"model":"embedl/efficienttam-ti-512","hardware":"l4","variant":"huggingface_fp16_tensorrt","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":4.835999999999999,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true}]},{"title":"Accuracy","subtitle":"DAVIS 2017 val · 8 sequences, 544 frames","metrics":[{"description":"J&F","value":81.88095696514159,"higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":583.1,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":59.730000000000004,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Video","Segmentation"],"build":"202611042314701a","updatedAt":"2026-11-10T15:54:34.000Z"}]},{"name":"timm/efficientvim_m1.e300_in1k","title":"EfficientViM-M1","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fefficientvim_m1.e300_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/efficientvim_m1.e300_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":478.76,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":479.614,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":2082,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":72.29,"unit":"%","headline":true},{"description":"Top-5","value":90.72,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/efficientvit_b0.r224_in1k","title":"EfficientViT-B0","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fefficientvit_b0.r224_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/efficientvit_b0.r224_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":333.099,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":333.801,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":2988.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.15,"unit":"%","headline":true},{"description":"Top-5","value":0.63,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/efficientvit_m0.r224_in1k","title":"EfficientViT-M0","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fefficientvit_m0.r224_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/efficientvit_m0.r224_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":835.876,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":836.304,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":1194.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":59.66,"unit":"%","headline":true},{"description":"Top-5","value":82.55,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientvit_m0.r224_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.904569,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.904907,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1103.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":59.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":82.54,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.55,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":30.3,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"a684af1915f8dc58","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/efficientvit_m1.r224_in1k","title":"EfficientViT-M1","source":"timm/efficientvit_m1.r224_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fefficientvit_m1.r224_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/efficientvit_m1.r224_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.727659,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.728149,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1371,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":65.23,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":86.47,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.56,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":38.2,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"885089644e76fcab","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/efficientvit_m2.r224_in1k","title":"EfficientViT-M2","source":"timm/efficientvit_m2.r224_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fefficientvit_m2.r224_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/efficientvit_m2.r224_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.759508,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.76001,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1313.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":67.91,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":87.95,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.57,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":52.8,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"0c8de5eed06ea718","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/efficientvit_m3.r224_in1k","title":"EfficientViT-M3","source":"timm/efficientvit_m3.r224_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fefficientvit_m3.r224_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/efficientvit_m3.r224_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.955856,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.974548,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1013.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":71.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.31,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4b9d495de64100bb","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientvit_m3.r224_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.04205,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.11078,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":322,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":71.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.31,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"850b19670d32bf8b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientvit_m3.r224_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.47512,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.48993,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":286.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":71.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.31,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3ae7b44e6055bc99","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientvit_m3.r224_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.833533,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.834045,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1197.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":71.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.31,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.6,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":85.4,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"4a5a21e029f13027","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientvit_m3.r224_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.35275,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.35522,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":738.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":71.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.31,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f21234b656de0e23","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/efficientvit_m4.r224_in1k","title":"EfficientViT-M4","source":"timm/efficientvit_m4.r224_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fefficientvit_m4.r224_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/efficientvit_m4.r224_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.977017,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.02826,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":994.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.95,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.2,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"75717ca51435d5da","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientvit_m4.r224_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.29562,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.35608,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":297.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.95,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.2,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"68cb7944e80a07a9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientvit_m4.r224_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.6499,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.67249,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":272.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.95,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.2,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8cefa34019218900","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientvit_m4.r224_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.895943,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.896454,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1114.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":72.95,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.2,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.63,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.11,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"052bf31b312352c7","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/efficientvit_m4.r224_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.42555,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.42767,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":701,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.95,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.2,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ee98d274ab74cc06","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/efficientvit_m5.r224_in1k","title":"EfficientViT-M5","source":"timm/efficientvit_m5.r224_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fefficientvit_m5.r224_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/efficientvit_m5.r224_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.12069,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.12134,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":891.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":74.92,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.12,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.67,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.15,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"b9e589a02b943284","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/ese_vovnet19b_dw.ra_in1k","title":"ESE-VoVNet-19-B-DW","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fese_vovnet19b_dw.ra_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/ese_vovnet19b_dw.ra_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":255.859,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":256.836,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":3881.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":74.79,"unit":"%","headline":true},{"description":"Top-5","value":92.21,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/ese_vovnet19b_dw.ra_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.295291,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.296326,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3370.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":74.92,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.32,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.6,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":75.8,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"9878954acfdf13cf","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/ese_vovnet39b.ra_in1k","title":"ESE-VoVNet-39-B","source":"timm/ese_vovnet39b.ra_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fese_vovnet39b.ra_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/ese_vovnet39b.ra_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.557325,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.639435,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1756.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.6,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3bda9415529b9730","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/ese_vovnet39b.ra_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.47585,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.48248,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":673.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.6,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"86289d3feca30835","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/ese_vovnet39b.ra_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.64019,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.64703,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":376.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.6,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"565384611792c272","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/ese_vovnet39b.ra_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.520986,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.673408,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.522278,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.676392,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1914,"unit":"qps","digits":1,"baseline":1481.55,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.6,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.78,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.26,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"3e7768e7bd9b7277","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/ese_vovnet39b.ra_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.17586,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.17926,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":849.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.6,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d430f248ab620ee6","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/ese_vovnet57b.ra4_e3600_r256_in1k","title":"ESE-VoVNet-57-B","source":"timm/ese_vovnet57b.ra4_e3600_r256_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fese_vovnet57b.ra4_e3600_r256_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/ese_vovnet57b.ra4_e3600_r256_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.767883,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.841492,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1279.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.29,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.91,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"04105f5f64a2d0a6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/ese_vovnet57b.ra4_e3600_r256_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.07735,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.33134,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":2.08514,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.34998,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":478.6,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":426.527,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.29,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.91,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"29363ab012d866de","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/ese_vovnet57b.ra4_e3600_r256_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.80919,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.82056,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":261.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.29,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.91,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"55d6056b47c4d63c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/ese_vovnet57b.ra4_e3600_r256_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.930962,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.11743,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.934448,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.12103,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1072.5,"unit":"qps","digits":1,"baseline":893.585,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.29,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.91,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.94,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.42,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"1ecd009048d822ec","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/ese_vovnet57b.ra4_e3600_r256_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.07148,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.08252,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":482.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.29,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.91,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ca7ffcb6d88c64f7","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/eva_giant_patch14_224.clip_ft_in1k","title":"EVA Giant/14 (224, CLIP, IN-1k fine-tuned)","source":"timm/eva_giant_patch14_224.clip_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Feva_giant_patch14_224.clip_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/eva_giant_patch14_224.clip_ft_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.45465,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":11.1024,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":8.60394,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":11.2152,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":117.4,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":89.4658,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":88.19,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.47,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2c719de4580662d8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva_giant_patch14_224.clip_ft_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":25.752,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":28.4886,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":25.8452,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":28.5778,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":38.6,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":34.9256,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":88.19,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.47,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"095a90242640bba8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva_giant_patch14_224.clip_ft_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1614.46,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":119,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":0.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":88.19,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.47,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"31aba4be7c2723be","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva_giant_patch14_224.clip_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":21.0253,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":21.7488,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":47.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":88.19,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.47,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":12.02,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":11.49,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"8aab9e78f9753f8d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva_giant_patch14_224.clip_ft_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":29.1126,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":29.5254,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":34.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":88.19,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.47,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"adf7fee2d722f264","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/eva_giant_patch14_clip_224.laion400m","title":"EVA Giant/14 (224, CLIP, LAION-400M)","source":"timm/eva_giant_patch14_clip_224.laion400m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Feva_giant_patch14_clip_224.laion400m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/eva_giant_patch14_clip_224.laion400m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":19.991,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":20.9862,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":50,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":9.43,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":8.91,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"bd3f4fef7797b80f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/eva_giant_patch14_336.clip_ft_in1k","title":"EVA Giant/14 (336, CLIP)","source":"timm/eva_giant_patch14_336.clip_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Feva_giant_patch14_336.clip_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/eva_giant_patch14_336.clip_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":39.9819,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":41.5574,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":25,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":88.59,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.53,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":8.02,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":7.5,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"ef379e00bf85091c","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/eva_giant_patch14_560.m30m_ft_in22k_in1k","title":"EVA Giant/14 (560)","source":"timm/eva_giant_patch14_560.m30m_ft_in22k_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Feva_giant_patch14_560.m30m_ft_in22k_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/eva_giant_patch14_560.m30m_ft_in22k_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":29.3667,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":44.5656,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":29.5549,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":47.9224,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":33.8,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":22.3244,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":89.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.75,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b212c3a1d26d4e08","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva_giant_patch14_560.m30m_ft_in22k_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":139.816,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":159.896,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":139.998,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":163.678,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":7.1,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":6.22287,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":89.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.75,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3f57ce5c6bca22b7","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva_giant_patch14_560.m30m_ft_in22k_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":219.501,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":226.066,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":4.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":89.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.75,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":12.68,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":12.16,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"11feeca179fb352e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva_giant_patch14_560.m30m_ft_in22k_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":210.749,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":213.605,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":4.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":89.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.75,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"61aa0cb89fe2003d","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/eva_large_patch14_196.in22k_ft_in1k","title":"EVA Large/14 (196)","source":"timm/eva_large_patch14_196.in22k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Feva_large_patch14_196.in22k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/eva_large_patch14_196.in22k_ft_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.51944,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.36436,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":2.61255,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.4884,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":393.2,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":293.961,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":86.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.97,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9355bc0146844482","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva_large_patch14_196.in22k_ft_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.43753,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":7.33409,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":6.45557,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":7.34814,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":154.5,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":135.643,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":86.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.97,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"dadc0b2da15c8ed8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva_large_patch14_196.in22k_ft_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":13.4405,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":13.4702,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":74,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":86.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.97,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a635f6909535f16d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva_large_patch14_196.in22k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.04649,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.4284,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":3.13782,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.50696,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":328.1,"unit":"qps","digits":1,"baseline":225.728,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":86.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.97,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.79,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.27,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"f81691e689fb9480","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva_large_patch14_196.in22k_ft_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.17678,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.2002,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":314.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":86.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.97,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9e5fc797aef47fff","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/eva_large_patch14_336.in22k_ft_in1k","title":"EVA Large/14 (336)","source":"timm/eva_large_patch14_336.in22k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Feva_large_patch14_336.in22k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/eva_large_patch14_336.in22k_ft_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.05216,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.6157,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":4.12292,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.70667,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":245.3,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":214.534,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":88.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.56,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"867ba81b76420801","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva_large_patch14_336.in22k_ft_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":13.6316,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":15.77,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":13.697,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":15.8321,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":73,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":63.0904,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":88.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.56,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7595824ce0ba66e7","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva_large_patch14_336.in22k_ft_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":33.1103,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":33.1876,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":30.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":88.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.56,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"26a51ab954ffe6de","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva_large_patch14_336.in22k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":14.1154,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":14.6653,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":70.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":88.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.56,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.91,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.39,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"c29e16b30d35ee72","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva_large_patch14_336.in22k_ft_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":14.3295,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":14.6119,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":69.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":88.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.56,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"67f7f0f3cbeb4309","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/eva02_base_patch14_224.mim_in22k","title":"EVA-02 Base/14 (224, MIM)","source":"timm/eva02_base_patch14_224.mim_in22k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Feva02_base_patch14_224.mim_in22k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/eva02_base_patch14_224.mim_in22k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.19246,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.26062,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":455.8,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.37,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.85,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"4cc99b3f4bbe1024","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/eva02_base_patch14_448.mim_in22k_ft_in1k","title":"EVA-02 Base/14 (448, MIM)","source":"timm/eva02_base_patch14_448.mim_in22k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Feva02_base_patch14_448.mim_in22k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/eva02_base_patch14_448.mim_in22k_ft_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.3352,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.40613,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":297.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":87.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.49,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"47c51740e0279c55","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva02_base_patch14_448.mim_in22k_ft_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.60898,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":11.657,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":9.62598,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":11.6838,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":103.5,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":85.349,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":87.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.49,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"271d66a42952f124","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva02_base_patch14_448.mim_in22k_ft_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":24.4502,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":24.4852,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":40.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":87.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.49,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"019db17b5bec2ecb","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva02_base_patch14_448.mim_in22k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":17.4806,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":17.9894,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":57.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":87.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.49,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"bd2615532d549d29","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva02_base_patch14_448.mim_in22k_ft_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":11.6209,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":11.7107,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":86,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":87.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.49,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"070563cf65c6861d","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/eva02_base_patch16_clip_224.merged2b","title":"EVA-02 Base/16 (224, CLIP)","source":"timm/eva02_base_patch16_clip_224.merged2b","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Feva02_base_patch16_clip_224.merged2b&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/eva02_base_patch16_clip_224.merged2b","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.52469,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.52795,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":655.2,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.38,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.86,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"d92ee0c243a1d4bb","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/eva02_enormous_patch14_clip_224.laion2b","title":"EVA-02 Enormous/14 (224, CLIP)","source":"timm/eva02_enormous_patch14_clip_224.laion2b","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Feva02_enormous_patch14_clip_224.laion2b&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/eva02_enormous_patch14_clip_224.laion2b","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":63.5347,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":73.8189,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":65.5715,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":77.5771,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":15.7,"unit":"qps","digits":1,"baseline":13.5464,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":34.42,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":33.9,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"9902ccaa2c25f0e2","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/eva02_large_patch14_clip_224.merged2b","title":"EVA-02 Large/14 (224, CLIP)","source":"timm/eva02_large_patch14_clip_224.merged2b","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Feva02_large_patch14_clip_224.merged2b&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/eva02_large_patch14_clip_224.merged2b","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.04985,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.21582,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":165.2,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.56,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.04,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"42e7b944351aa0ac","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/eva02_large_patch14_224.mim_in22k","title":"EVA-02 Large/14 (224, MIM)","source":"timm/eva02_large_patch14_224.mim_in22k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Feva02_large_patch14_224.mim_in22k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/eva02_large_patch14_224.mim_in22k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.5253,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.75165,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":153.2,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.52,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"ef659fa831166069","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/eva02_large_patch14_clip_336.merged2b","title":"EVA-02 Large/14 (336, CLIP)","source":"timm/eva02_large_patch14_clip_336.merged2b","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Feva02_large_patch14_clip_336.merged2b&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/eva02_large_patch14_clip_336.merged2b","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":16.3404,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":17.0612,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":61.2,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.56,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.04,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"ec317f673991938b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/eva02_large_patch14_448.mim_in22k_ft_in1k","title":"EVA-02 Large/14 (448, MIM)","source":"timm/eva02_large_patch14_448.mim_in22k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Feva02_large_patch14_448.mim_in22k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/eva02_large_patch14_448.mim_in22k_ft_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":10.6376,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":12.4828,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":10.7976,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":12.6118,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":93.3,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":79.9354,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":89.39,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.89,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"11874e4400e68ac0","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva02_large_patch14_448.mim_in22k_ft_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":34.6088,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":41.4558,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":34.7151,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":41.521,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":28.7,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":24.0014,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":89.39,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.89,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3a903f934a02445e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva02_large_patch14_448.mim_in22k_ft_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":86.2876,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":86.7429,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":11.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":89.39,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.89,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0ded739757706aac","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva02_large_patch14_448.mim_in22k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":50.5114,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":52.4369,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":19.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":89.39,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.89,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.4,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"4cbf504767a69d6f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva02_large_patch14_448.mim_in22k_ft_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":35.9408,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":36.1543,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":27.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":89.39,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.89,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"523bc44cf9b7b2a8","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/eva02_small_patch14_224.mim_in22k","title":"EVA-02 Small/14 (224, MIM)","source":"timm/eva02_small_patch14_224.mim_in22k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Feva02_small_patch14_224.mim_in22k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/eva02_small_patch14_224.mim_in22k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.07561,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.08466,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":928.3,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.72,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.2,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"71bd0443b23653af","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/eva02_small_patch14_336.mim_in22k_ft_in1k","title":"EVA-02 Small/14 (336, MIM)","source":"timm/eva02_small_patch14_336.mim_in22k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Feva02_small_patch14_336.mim_in22k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/eva02_small_patch14_336.mim_in22k_ft_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.09302,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.17682,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":895.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.95,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.95,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f605c785aa2c5704","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva02_small_patch14_336.mim_in22k_ft_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.50601,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.51501,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":396.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.95,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.95,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c9a8d1a2ebe6cb5c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva02_small_patch14_336.mim_in22k_ft_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.8441,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.85779,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":170.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.95,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.95,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3e1d50dd93871e2d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva02_small_patch14_336.mim_in22k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.38642,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.46796,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":418.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.95,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.95,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.78,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.26,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"b3d7ba4ad2ccdeff","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva02_small_patch14_336.mim_in22k_ft_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.11552,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.15662,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":320.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.95,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.95,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"087c3e56b274f16a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/eva02_tiny_patch14_224.mim_in22k","title":"EVA-02 Tiny/14 (224, MIM)","source":"timm/eva02_tiny_patch14_224.mim_in22k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Feva02_tiny_patch14_224.mim_in22k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/eva02_tiny_patch14_224.mim_in22k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.614143,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.616394,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1623.9,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.57,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":52.3,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image"],"build":"57db440e02c05276","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/eva02_tiny_patch14_336.mim_in22k_ft_in1k","title":"EVA-02 Tiny/14 (336, MIM)","source":"timm/eva02_tiny_patch14_336.mim_in22k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Feva02_tiny_patch14_336.mim_in22k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/eva02_tiny_patch14_336.mim_in22k_ft_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.810166,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.892029,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1199.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.95,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d3c2096f3cc3aaed","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva02_tiny_patch14_336.mim_in22k_ft_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.38403,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.39008,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":718.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.95,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"08c20b89ac3db8d8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva02_tiny_patch14_336.mim_in22k_ft_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.91377,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.92432,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":341.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.95,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4a82b2a51e352bc7","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva02_tiny_patch14_336.mim_in22k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.2311,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.24219,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":811.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.95,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":71.9,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"00ef5dddabbf2452","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/eva02_tiny_patch14_336.mim_in22k_ft_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.51263,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.52002,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":660.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.95,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"238ddcb67a04bdbe","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/fasternet_l.in1k","title":"FasterNet-L","source":"timm/fasternet_l.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffasternet_l.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/fasternet_l.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.11166,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.38571,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1.20453,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.49487,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":880.6,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":708.445,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.21,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.56,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9feb34098f5a0ce0","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fasternet_l.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.59349,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.6073,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":276.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.21,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.56,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9cf0b1b587d47a31","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fasternet_l.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.76116,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.78674,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":147.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.21,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.56,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0b83b74ddca77008","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fasternet_l.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.55425,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.56079,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":642.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.21,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.56,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.62,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.1,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"a1c1670a371b51dc","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fasternet_l.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.20538,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.21509,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":453.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.21,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.56,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"07c18188e5d8d266","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/fasternet_m.in1k","title":"FasterNet-M","source":"timm/fasternet_m.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffasternet_m.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/fasternet_m.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.914453,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.06719,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1.00568,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.15643,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1074.2,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":916.946,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.15,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"afed27b8ec6c07c8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fasternet_m.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.17103,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.1853,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":313.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.15,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"98c9c27b99ba4d93","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fasternet_m.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.00045,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.02386,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":165.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.15,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2c2efda1ca0c2d4e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fasternet_m.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.25821,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.26141,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":793.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.15,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.15,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.63,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"dbe83ea52410293a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fasternet_m.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.8984,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.90594,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":526.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.15,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0ddc94e418aaa9e3","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/fasternet_s.in1k","title":"FasterNet-S","source":"timm/fasternet_s.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffasternet_s.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/fasternet_s.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.521806,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.606517,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.598145,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.679443,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1865.1,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1603.03,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.65,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.29,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"30561f8dce0dfa95","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fasternet_s.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.72869,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.73431,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":575.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.65,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.29,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d6212d91d8595d6c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fasternet_s.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.96612,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.97382,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":335.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.65,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.29,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ebf25dd1c2f6186b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fasternet_s.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.712152,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.715454,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1401.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.65,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.29,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.89,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.37,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"bd4546c962192995","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fasternet_s.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.16524,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.16711,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":857.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.65,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.29,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3e0df2ae4d3b5c00","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/fasternet_t0.in1k","title":"FasterNet-T0","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffasternet_t0.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/fasternet_t0.in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":380.981,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":381.897,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":2614.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":36.36,"unit":"%","headline":true},{"description":"Top-5","value":61.39,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/fasternet_t2.in1k","title":"FasterNet-T2","source":"timm/fasternet_t2.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffasternet_t2.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/fasternet_t2.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.308709,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.381531,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3190.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.45,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.83,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"678d159cc87ffcfb","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fasternet_t2.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.25892,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.26544,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":789.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.45,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.83,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0f7d489eea072e36","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fasternet_t2.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.00078,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.00671,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":496.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.45,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.83,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"36016a2f22b3c9fc","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fasternet_t2.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.455853,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.457092,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2187.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.45,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.83,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.7,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.18,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"8c9da3022c229d51","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fasternet_t2.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.793727,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.795898,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1258.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.45,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.83,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5548e79858e07623","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/fastvit_ma36.apple_dist_in1k","title":"FastViT-MA36","source":"timm/fastvit_ma36.apple_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffastvit_ma36.apple_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/fastvit_ma36.apple_dist_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.20141,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.30585,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":310.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.96,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ad34ce507d7fcbfa","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_ma36.apple_dist_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":22.2334,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":22.2715,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":45,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.96,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"089f945010f3ba58","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_ma36.apple_dist_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":34.4548,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":34.5105,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":29,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.96,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d45c6e499032f95f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_ma36.apple_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.73797,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":9.74634,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":102.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.96,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.13,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.61,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"be4b90c25a0be5c5","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_ma36.apple_dist_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":16.791,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":16.8092,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":59.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.96,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7b7c21c72e6f5a9b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/fastvit_mci0.apple_mclip","title":"FastViT-MCI0","source":"timm/fastvit_mci0.apple_mclip","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffastvit_mci0.apple_mclip&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/fastvit_mci0.apple_mclip","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.03851,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":7.04321,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":142,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.68,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.16,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"b44875f128bf9b72","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/fastvit_mci1.apple_mclip","title":"FastViT-MCI1","source":"timm/fastvit_mci1.apple_mclip","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffastvit_mci1.apple_mclip&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/fastvit_mci1.apple_mclip","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.25443,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":8.25952,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":121.1,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.82,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.3,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"c050a45aefa8be24","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/fastvit_mci2.apple_mclip","title":"FastViT-MCI2","source":"timm/fastvit_mci2.apple_mclip","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffastvit_mci2.apple_mclip&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/fastvit_mci2.apple_mclip","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":10.1047,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":10.1113,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":98.9,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.02,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.5,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"762f7980123de539","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/fastvit_mci3.apple_mclip2_dfndr2b","title":"FastViT-MCI3","source":"timm/fastvit_mci3.apple_mclip2_dfndr2b","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffastvit_mci3.apple_mclip2_dfndr2b&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/fastvit_mci3.apple_mclip2_dfndr2b","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":21.7404,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":21.7484,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":46,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.18,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.66,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"29c3cf1dfb4f3cd1","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/fastvit_mci4.apple_mclip2_dfndr2b","title":"FastViT-MCI4","source":"timm/fastvit_mci4.apple_mclip2_dfndr2b","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffastvit_mci4.apple_mclip2_dfndr2b&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/fastvit_mci4.apple_mclip2_dfndr2b","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":31.4435,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":31.4736,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":31.8,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4.75,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":4.23,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"1dbac4895df848bc","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/fastvit_s12.apple_dist_in1k","title":"FastViT-S12","source":"timm/fastvit_s12.apple_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffastvit_s12.apple_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/fastvit_s12.apple_dist_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.10033,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.19739,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":890.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.66,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.12,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b25160c09fdf0010","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_s12.apple_dist_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":15.3583,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":15.3867,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":65.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.66,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.12,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"73d4110ce660905f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_s12.apple_dist_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":21.7461,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":21.8016,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":45.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.66,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.12,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"51687248d886cefa","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_s12.apple_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.66666,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.67126,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":150,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.66,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.12,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.65,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.13,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"de2f4661ebe9bd70","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_s12.apple_dist_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":11.6296,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":11.6388,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":86,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.66,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.12,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4d080a497ddb0262","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/fastvit_sa12.apple_dist_in1k","title":"FastViT-SA12","source":"timm/fastvit_sa12.apple_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffastvit_sa12.apple_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/fastvit_sa12.apple_dist_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.16238,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.25995,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":844.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.23,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.41,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f2affcb71ea4d4ea","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_sa12.apple_dist_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":15.3514,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":15.3776,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":65.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.23,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.41,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a4dfdc9ec75cbe38","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_sa12.apple_dist_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":21.7538,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":21.8146,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":45.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.23,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.41,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4b850e83f25c93e3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_sa12.apple_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.71065,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.71619,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":149,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.23,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.41,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.68,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.16,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"2a3c68d9132668f5","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_sa12.apple_dist_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":11.6638,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":11.6699,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":85.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.23,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.41,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0b01a0a88788b539","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/fastvit_sa24.apple_dist_in1k","title":"FastViT-SA24","source":"timm/fastvit_sa24.apple_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffastvit_sa24.apple_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/fastvit_sa24.apple_dist_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.07529,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.19684,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":475,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.37,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d0ff06ad931b65e6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_sa24.apple_dist_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":16.9305,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":16.9644,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":59,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.37,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"17e982cb5444a9f7","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_sa24.apple_dist_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":24.7294,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":24.7969,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":40.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.37,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e3173fdd11e7fbbe","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_sa24.apple_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.46655,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":7.47192,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":133.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.37,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.82,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.3,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"1a987ea8c3080908","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_sa24.apple_dist_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":12.6866,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":12.708,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":78.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.37,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a49846c75ef4f2e0","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/fastvit_sa36.apple_dist_in1k","title":"FastViT-SA36","source":"timm/fastvit_sa36.apple_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffastvit_sa36.apple_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/fastvit_sa36.apple_dist_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.85215,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.96283,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":346.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.91,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.8,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9ee55a9ba927dc93","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_sa36.apple_dist_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":18.4727,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":18.5176,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":54.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.91,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.8,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"da074f89abdcf63f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_sa36.apple_dist_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":27.7901,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":27.8806,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":36,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.91,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.8,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e8e2df4ac29bf556","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_sa36.apple_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.11381,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":8.12231,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":123.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.91,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.8,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.96,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.44,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"c18d40154c5ef2cc","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_sa36.apple_dist_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":13.5431,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":13.6218,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":73.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.91,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.8,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3d3b9f218e395c03","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/fastvit_t12.apple_dist_in1k","title":"FastViT-T12","source":"timm/fastvit_t12.apple_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffastvit_t12.apple_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/fastvit_t12.apple_dist_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.08911,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.17993,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":897.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.86,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.76,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"78a27c11ad20e4b4","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_t12.apple_dist_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":15.216,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":15.2485,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":65.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.86,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.76,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"db61f8f2a79fc2b5","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_t12.apple_dist_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":21.5682,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":21.6168,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":46.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.86,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.76,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e36bc8aa66149f45","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_t12.apple_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.69894,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.70306,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":149.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.86,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.76,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.63,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.11,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"728db37c37c8331f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_t12.apple_dist_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":11.3677,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":11.376,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":88,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.86,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.76,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f83aa793e6bc2189","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/fastvit_t8.apple_dist_in1k","title":"FastViT-T8","source":"timm/fastvit_t8.apple_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffastvit_t8.apple_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/fastvit_t8.apple_dist_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.871994,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.956116,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1117.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.31,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.81,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d523e25235e7b071","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_t8.apple_dist_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":11.5778,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":11.5981,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":86.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.31,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.81,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4b8aa00182ee47eb","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_t8.apple_dist_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":16.2774,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":16.3237,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":61.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.31,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.81,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0752683ec9999058","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_t8.apple_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.74314,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.74579,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":210.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.31,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.81,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.58,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":58.1,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"6166dd4425f521f9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/fastvit_t8.apple_dist_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.62641,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":8.63257,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":115.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.31,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.81,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f280081389098592","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/fbnetc_100.rmsp_in1k","title":"FBNet-C 1.0×","source":"timm/fbnetc_100.rmsp_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffbnetc_100.rmsp_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/fbnetc_100.rmsp_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.224681,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.225281,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":4419.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":72.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.18,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":65.6,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"b3afbd98279adfba","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/fbnetv3_b.ra2_in1k","title":"FBNetV3-B","source":"timm/fbnetv3_b.ra2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffbnetv3_b.ra2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/fbnetv3_b.ra2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.941909,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.942993,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1060,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.23,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.63,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.11,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"2dcfb06eab57d526","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/fbnetv3_d.ra2_in1k","title":"FBNetV3-D","source":"timm/fbnetv3_d.ra2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffbnetv3_d.ra2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/fbnetv3_d.ra2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.00999,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.01111,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":988.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.4,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.31,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.65,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.13,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e4f4d70b7eb7fe7e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/fcn_resnet101","title":"FCN ResNet-101","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Ffcn_resnet101&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/fcn_resnet101","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":14.139,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":14.546,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":70.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"COCO segmentation validation · 2,000 images · Baseline TensorRT as baseline","metrics":[{"description":"mIoU","value":64,"unit":"%","headline":true}]}],"labels":["Image","Segmentation"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/fcn_resnet101","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":14.139,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":14.546,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":70.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"COCO segmentation validation · 2,000 images · Baseline TensorRT as baseline","metrics":[{"description":"mIoU","value":64,"unit":"%","headline":true}]}],"labels":["Image","Segmentation"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/fcn_resnet50","title":"FCN ResNet-50","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Ffcn_resnet50&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/fcn_resnet50","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":8.267,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":8.59,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":120.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"COCO segmentation validation · 2,000 images · Baseline TensorRT as baseline","metrics":[{"description":"mIoU","value":59.65,"unit":"%","headline":true}]}],"labels":["Image","Segmentation"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/fcn_resnet50","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":8.267,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":8.59,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":120.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"COCO segmentation validation · 2,000 images · Baseline TensorRT as baseline","metrics":[{"description":"mIoU","value":59.65,"unit":"%","headline":true}]}],"labels":["Image","Segmentation"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/flexivit_base.300ep_in1k","title":"FlexiViT Base","source":"timm/flexivit_base.300ep_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fflexivit_base.300ep_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/flexivit_base.300ep_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.10577,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.20126,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":883.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.35,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.61,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"cd3be568f6a85802","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/flexivit_base.300ep_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.45222,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.96804,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":2.46124,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.00293,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":405.5,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":335.097,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.35,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.61,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"197ca69b363a4e47","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/flexivit_base.300ep_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.22337,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.23621,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":190.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.35,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.61,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"06d247826b38249d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/flexivit_base.300ep_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.978574,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.34226,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.981384,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.34589,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1020.4,"unit":"qps","digits":1,"baseline":744.1,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.35,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.61,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.49,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.97,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"8b1ccc4a95a0f6b8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/flexivit_base.300ep_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.38195,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.38879,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":723,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.35,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.61,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3b59d0185765a63b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/flexivit_large.300ep_in1k","title":"FlexiViT Large","source":"timm/flexivit_large.300ep_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fflexivit_large.300ep_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/flexivit_large.300ep_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.61442,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.41172,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":2.70569,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.51666,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":377.7,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":289.817,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.9,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a2588d7da94927e6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/flexivit_large.300ep_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.7454,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":7.98496,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":6.76337,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":7.99597,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":147.5,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":124.589,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.9,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f8993c8dd3340a37","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/flexivit_large.300ep_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":13.9865,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":14.1527,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":71.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.9,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5495e1b8e7c35385","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/flexivit_large.300ep_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.29811,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.64859,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":3.36627,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.77344,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":303.1,"unit":"qps","digits":1,"baseline":215.044,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.9,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.93,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.41,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"625d3a778b52a903","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/flexivit_large.300ep_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.1691,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.18762,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":315.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.9,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9204d69044207a69","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"embedl/flux-3-action-droid","upstream":"black-forest-labs/flux-3-action-droid","title":"FLUX 3 Action DROID","source":"embedl/flux-3-action-droid","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=embedl%2Fflux-3-action-droid&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"embedl/flux-3-action-droid","hardware":"thor","precision":"FP8","metrics":[{"title":"Latency","subtitle":"One DiT step (one 32-action chunk) · batch 1 · trtexec · text encoder and VAE not included","metrics":[{"description":"DiT step, median","value":209.82,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"DiT step, p99","value":222.67,"unit":"ms","higherIsBetter":false,"digits":2}]},{"title":"Accuracy","subtitle":"RoboLab-120 · 120 tasks × 10 episodes · published by Black Forest Labs for variants/sd","metrics":[{"description":"Task success on RoboLab-120 (published)","value":37.92,"unit":"%","higherIsBetter":true,"digits":2,"headline":true},{"description":"Action output vs PyTorch, max abs","value":0.046875,"unit":"","higherIsBetter":false,"digits":4}]}],"labels":["Text","VLA","Robot Policy"],"build":"202610201828089b","updatedAt":"2026-10-29T16:07:23.000Z","simulatorReachable":false},{"model":"embedl/flux-3-action-droid","hardware":"h200","precision":"FP8","metrics":[{"title":"Latency","subtitle":"One DiT step (one 32-action chunk) · batch 1 · trtexec · text encoder and VAE not included","metrics":[{"description":"DiT step, median","value":38.37,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"DiT step, p99","value":39.42,"unit":"ms","higherIsBetter":false,"digits":2}]},{"title":"Accuracy","subtitle":"RoboLab-120 · 120 tasks × 10 episodes · published by Black Forest Labs for variants/sd","metrics":[{"description":"Task success on RoboLab-120 (published)","value":37.92,"unit":"%","higherIsBetter":true,"digits":2,"headline":true},{"description":"Action output vs PyTorch, max abs","value":0.09375,"unit":"","higherIsBetter":false,"digits":4}]}],"labels":["Text","VLA","Robot Policy"],"build":"2026101814036bb4","updatedAt":"2026-10-29T16:07:23.000Z","simulatorReachable":false}]},{"name":"black-forest-labs/flux-2-klein-4b","upstream":"black-forest-labs/FLUX.2-klein-4B","title":"FLUX.2 [klein] 4B","source":"black-forest-labs/flux-2-klein-4b","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=black-forest-labs%2FFLUX.2-klein-4B&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"black-forest-labs/flux-2-klein-4b","hardware":"thor","precision":"FP8","metrics":[{"title":"Latency","metrics":[{"description":"Median time per image","value":430.5658499997662,"unit":"ms","higherIsBetter":false,"digits":2,"baseline":639.9425125564449,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Accuracy","subtitle":"PartiPrompts · 200 prompts","metrics":[{"description":"CLIPScore","value":32.10484313964844,"unit":"score","higherIsBetter":true,"digits":3,"baseline":32.060813903808594,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory, serving process","value":3327.7734375,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":4913.4296875,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Peak memory, TensorRT","value":10.525327095761895,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":13.975535752251744,"baselineLabel":"Baseline TensorRT"}]},{"title":"Size","metrics":[{"description":"Model size","value":9.7,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":13.12,"baselineLabel":"Baseline TensorRT","headline":true}]}],"labels":["Text","Image","Generation"],"build":"2026101817525bbe","updatedAt":"2026-11-10T15:54:34.000Z"},{"model":"black-forest-labs/flux-2-klein-4b","hardware":"thor","variant":"huggingface_bf16_tensorrt","precision":"BF16","metrics":[{"title":"Latency","metrics":[{"description":"Median time per image","value":639.9425125564449,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true}]},{"title":"Accuracy","subtitle":"PartiPrompts · 200 prompts","metrics":[{"description":"CLIPScore","value":32.060813903808594,"unit":"score","higherIsBetter":true,"digits":3,"headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory, serving process","value":4913.4296875,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true},{"description":"Peak memory, TensorRT","value":13.975535752251744,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Model size","value":13.12,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true}]}],"labels":["Text","Image","Generation"],"build":"202610291630bdd4","updatedAt":"2026-11-10T15:54:34.000Z"},{"model":"black-forest-labs/flux-2-klein-4b","hardware":"thor","variant":"huggingface_fp8_embedl_tensorrt","precision":"FP8","metrics":[{"title":"Latency","metrics":[{"description":"Median time per image","value":430.5658499997662,"unit":"ms","higherIsBetter":false,"digits":2,"baseline":639.9425125564449,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Accuracy","subtitle":"PartiPrompts · 200 prompts","metrics":[{"description":"CLIPScore","value":32.10484313964844,"unit":"score","higherIsBetter":true,"digits":3,"baseline":32.060813903808594,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Memory","metrics":[{"description":"Peak memory, serving process","value":3327.7734375,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":4913.4296875,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Peak memory, TensorRT","value":10.525327095761895,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":13.975535752251744,"baselineLabel":"Baseline TensorRT"}]},{"title":"Size","metrics":[{"description":"Model size","value":9.7,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":13.12,"baselineLabel":"Baseline TensorRT","headline":true}]}],"labels":["Text","Image","Generation"],"build":"2026101817525bbe","updatedAt":"2026-11-10T15:54:34.000Z"},{"model":"black-forest-labs/flux-2-klein-4b","hardware":"h200","precision":"FP8","metrics":[{"title":"Latency","subtitle":"512×512 · batch 1 · 4 steps · 200 prompts, warm","metrics":[{"description":"Prompt to PNG, median","value":126.38319400139153,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"Prompt to PNG, p95","value":133.7185080628842,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Text encoding","value":11.795526486821473,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Four denoise steps","value":74.59741400089115,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"VAE and postprocess","value":15.443067415617406,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"PNG and base64","value":24.12347251083702,"unit":"ms","higherIsBetter":false,"digits":2}]},{"title":"Accuracy","subtitle":"PartiPrompts · 200 paired fixed-seed prompts · upstream BF16 as baseline","metrics":[{"description":"CLIPScore","value":32.04734802246094,"unit":"score","higherIsBetter":true,"digits":3,"baseline":32.16584396362305,"baselineLabel":"Upstream BF16","headline":true},{"description":"CLIP-FID against BF16","value":3.862588872467171,"unit":"distance","higherIsBetter":false,"digits":4}]},{"title":"Memory","metrics":[{"description":"Peak serving process RSS","value":3235.55078125,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true},{"description":"TensorRT allocator and I/O buffers, peak requested bytes","value":10.652092019096017,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","Image","Generation"],"build":"202610181752280b","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"black-forest-labs/flux-2-klein-4b","hardware":"h200","variant":"huggingface_bf16_pytorch","precision":"BF16","metrics":[{"title":"Latency","subtitle":"1024×1024 · batch 1 · 4 steps · 198 PartiPrompts, warm","metrics":[{"description":"PartiPrompts, 198 warm prompts, median","value":900.1335,"unit":"ms","higherIsBetter":false,"digits":0,"headline":true},{"description":"Prompt to image, warm median (1024², 4 steps)","value":851.082,"unit":"ms","higherIsBetter":false,"digits":0},{"description":"Prompt to image, warm p95","value":929.8223,"unit":"ms","higherIsBetter":false,"digits":0},{"description":"Text encoding, median","value":27.729,"unit":"ms","higherIsBetter":false,"digits":1},{"description":"Four denoise steps, median","value":365.881,"unit":"ms","higherIsBetter":false,"digits":1},{"description":"VAE decode and postprocess, median","value":107.4125,"unit":"ms","higherIsBetter":false,"digits":1},{"description":"PNG and base64 encoding, median","value":350.79449999999997,"unit":"ms","higherIsBetter":false,"digits":1},{"description":"PartiPrompts, warm p95","value":1040.9968,"unit":"ms","higherIsBetter":false,"digits":0}]},{"title":"Accuracy","subtitle":"PartiPrompts · 200 fixed-seed prompts at 1024² · independent upstream BF16 reference","metrics":[{"description":"CLIPScore, served build","value":31.98045862197876,"unit":"score","higherIsBetter":true,"digits":3,"headline":true},{"description":"CLIPScore, BF16 reference","value":31.98045862197876,"unit":"score","higherIsBetter":true,"digits":3},{"description":"Paired CLIPScore delta","value":0,"unit":"score","higherIsBetter":true,"digits":4},{"description":"Images matching reference pixels exactly","value":200,"unit":"/200","higherIsBetter":true,"digits":0}]},{"title":"Memory","metrics":[{"description":"Peak GPU tensor allocation, 1024²","value":17704.0068359375,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true},{"description":"Peak GPU allocator reservation, 1024²","value":20024,"unit":"MiB","higherIsBetter":false,"digits":0},{"description":"Peak process RSS","value":8583.55078125,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Text","Image","Generation"],"build":"202610180925d9c4","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/focalnet_base_lrf.ms_in1k","title":"FocalNet Base (lrf)","source":"timm/focalnet_base_lrf.ms_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffocalnet_base_lrf.ms_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/focalnet_base_lrf.ms_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.95618,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.08459,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":335.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.44,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.49,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"12b4846fa889efff","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/focalnet_base_lrf.ms_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":11.7921,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":11.8154,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":84.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.44,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.49,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b60369469bc6f9fb","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/focalnet_base_lrf.ms_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":20.7688,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":20.7952,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":48,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.44,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.49,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3dacc21679abdfd0","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/focalnet_base_lrf.ms_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.00221,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.07764,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":249.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.44,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.49,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.52,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"70361985558ae69f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/focalnet_base_lrf.ms_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.46833,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.48248,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":182.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.44,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.49,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6919ec9abdb21001","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/focalnet_base_srf.ms_in1k","title":"FocalNet Base (srf)","source":"timm/focalnet_base_srf.ms_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffocalnet_base_srf.ms_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/focalnet_base_srf.ms_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.46297,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.59515,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":403.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.53,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"394d021c3bf20822","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/focalnet_base_srf.ms_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.96869,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":9.99341,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":99.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.53,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"75f4ab4accbdf311","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/focalnet_base_srf.ms_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":17.5622,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":17.5848,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":56.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.53,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"24f38d587dd38678","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/focalnet_base_srf.ms_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.26727,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.32678,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":305.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.53,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"92a348291930d575","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/focalnet_base_srf.ms_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.61182,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.63062,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":216.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.53,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f4e16d44e80eac41","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/focalnet_huge_fl3.ms_in22k","title":"FocalNet Huge (fl3)","source":"timm/focalnet_huge_fl3.ms_in22k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffocalnet_huge_fl3.ms_in22k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/focalnet_huge_fl3.ms_in22k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":12.2267,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":12.4662,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":81.8,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":8.17,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":7.65,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"da87dd448cb89e23","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/focalnet_huge_fl4.ms_in22k","title":"FocalNet Huge (fl4)","source":"timm/focalnet_huge_fl4.ms_in22k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffocalnet_huge_fl4.ms_in22k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/focalnet_huge_fl4.ms_in22k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":16.2722,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":16.5188,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":61.4,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":5.67,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":5.15,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"e86bbb8caba3f84d","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/focalnet_large_fl3.ms_in22k","title":"FocalNet Large (fl3)","source":"timm/focalnet_large_fl3.ms_in22k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffocalnet_large_fl3.ms_in22k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/focalnet_large_fl3.ms_in22k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":17.9779,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":18.4849,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":55.6,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.07,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.55,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"60b863857236e47b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/focalnet_large_fl4.ms_in22k","title":"FocalNet Large (fl4)","source":"timm/focalnet_large_fl4.ms_in22k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffocalnet_large_fl4.ms_in22k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/focalnet_large_fl4.ms_in22k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":18.8117,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":19.2051,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":53.2,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.07,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.55,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"0524f3a5f8ad9c35","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/focalnet_small_lrf.ms_in1k","title":"FocalNet Small (lrf)","source":"timm/focalnet_small_lrf.ms_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffocalnet_small_lrf.ms_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/focalnet_small_lrf.ms_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.68301,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.85321,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":370.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.43,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"873ece2919c91ab0","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/focalnet_small_lrf.ms_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.84319,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":9.85986,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":101.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.43,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d953f3d5a751c72e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/focalnet_small_lrf.ms_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":18.0246,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":18.0511,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":55.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.43,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2cf6d48a83c0d447","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/focalnet_small_lrf.ms_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.12912,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.13367,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":319.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.43,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.09,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.57,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"9dd9aea5ab3f6008","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/focalnet_small_lrf.ms_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.55957,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.5788,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":219.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.43,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d877f6d22a6a4739","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/focalnet_small_srf.ms_in1k","title":"FocalNet Small (srf)","source":"timm/focalnet_small_srf.ms_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffocalnet_small_srf.ms_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/focalnet_small_srf.ms_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.21405,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.34778,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":446.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.29,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"faaf47e17a45c7b3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/focalnet_small_srf.ms_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.68777,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":8.71936,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":114.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.29,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7b7a0730542453ce","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/focalnet_small_srf.ms_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":15.1509,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":15.1743,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":65.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.29,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"bc8cf8fdb2e22096","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/focalnet_small_srf.ms_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.66792,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.67157,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":374.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.29,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.08,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.56,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"9544b4c7acbca8f9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/focalnet_small_srf.ms_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.88624,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.89124,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":257.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.29,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f9945b0c599be145","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/focalnet_tiny_lrf.ms_in1k","title":"FocalNet Tiny (lrf)","source":"timm/focalnet_tiny_lrf.ms_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffocalnet_tiny_lrf.ms_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/focalnet_tiny_lrf.ms_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.50658,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.62268,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":656.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.74,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.82,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"21a26d5ce908fe42","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/focalnet_tiny_lrf.ms_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.31233,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.33441,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":187.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.74,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.82,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"276ad69a58948fe6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/focalnet_tiny_lrf.ms_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.83336,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":9.85089,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":101.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.74,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.82,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9c573a92364b6f77","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/focalnet_tiny_lrf.ms_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.57358,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.57642,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":634.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.74,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.82,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.84,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.32,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"2d5d435e2ea64498","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/focalnet_tiny_lrf.ms_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.46702,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.47192,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":405.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.74,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.82,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"63523efb790eced1","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/focalnet_tiny_srf.ms_in1k","title":"FocalNet Tiny (srf)","source":"timm/focalnet_tiny_srf.ms_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffocalnet_tiny_srf.ms_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/focalnet_tiny_srf.ms_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.25556,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.37665,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":784.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.83,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d853a97afd1edd76","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/focalnet_tiny_srf.ms_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.51829,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.53638,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":220.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.83,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a6b01b6acd399870","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/focalnet_tiny_srf.ms_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.16411,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":8.1781,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":121.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.83,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"effdcfb8929c1e91","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/focalnet_tiny_srf.ms_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.32768,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.32953,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":752.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.83,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.84,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.32,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"292592e05e1c9a15","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/focalnet_tiny_srf.ms_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.05163,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.0553,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":487.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.83,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"77fa1a005457d4f5","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/focalnet_xlarge_fl3.ms_in22k","title":"FocalNet XLarge (fl3)","source":"timm/focalnet_xlarge_fl3.ms_in22k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffocalnet_xlarge_fl3.ms_in22k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/focalnet_xlarge_fl3.ms_in22k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":27.3563,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":27.8226,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":36.6,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.26,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.74,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"17955c55609cec70","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/focalnet_xlarge_fl4.ms_in22k","title":"FocalNet XLarge (fl4)","source":"timm/focalnet_xlarge_fl4.ms_in22k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ffocalnet_xlarge_fl4.ms_in22k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/focalnet_xlarge_fl4.ms_in22k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":28.9596,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":29.4702,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":34.5,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4.6,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":4.08,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"66c16cda589db9bc","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"embedl/bevformer-base","upstream":"fundamentalvision/BEVFormer","upstreamUrl":"https://github.com/fundamentalvision/BEVFormer","source":"fundamentalvision/BEVFormer","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=embedl%2Fbevformer-base&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"embedl/bevformer-base","hardware":"orin","variant":"fp16","precision":"FP16","metrics":[{"title":"Latency","subtitle":"batch 1 · one nuScenes sample · GPU compute only","metrics":[{"description":"GPU compute, mean","value":1150.31,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"GPU compute, p99","value":1151.56,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":0.87,"unit":"fps","higherIsBetter":true,"digits":2}]},{"title":"Accuracy","subtitle":"nuScenes val · 6,019 samples · PyTorch FP16 on an NVIDIA L4, not this engine","metrics":[{"description":"NDS (nuScenes val)","value":51.7,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"mAP (nuScenes val)","value":41.6,"unit":"%","higherIsBetter":true,"digits":1}]}],"labels":["Image","Detection"],"build":"20261013092191dc","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"embedl/bevformer-base","hardware":"orin","variant":"int8","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · one nuScenes sample · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"GPU compute, mean","value":1143.06,"unit":"ms","higherIsBetter":false,"digits":2,"baseline":1150.31,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1143.77,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":0.87,"unit":"fps","higherIsBetter":true,"digits":2}]},{"title":"Accuracy","subtitle":"nuScenes val · 6,019 samples · PyTorch FP16 with INT8 fake quantization on an NVIDIA L4, not this engine","metrics":[{"description":"NDS (nuScenes val)","value":51.1,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"mAP (nuScenes val)","value":40.9,"unit":"%","higherIsBetter":true,"digits":1}]}],"labels":["Image","Detection"],"build":"2026101309217d8b","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"embedl/bevformer-base","hardware":"l4","variant":"fp16","precision":"FP16","metrics":[{"title":"Latency","subtitle":"batch 1 · one nuScenes sample · GPU compute only","metrics":[{"description":"GPU compute, mean","value":455.13,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"GPU compute, p99","value":461.08,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":2.2,"unit":"fps","higherIsBetter":true,"digits":2}]},{"title":"Accuracy","subtitle":"nuScenes val · 6,019 samples · PyTorch FP16 on an NVIDIA L4, not this engine","metrics":[{"description":"NDS (nuScenes val)","value":51.7,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"mAP (nuScenes val)","value":41.6,"unit":"%","higherIsBetter":true,"digits":1}]}],"labels":["Image","Detection"],"build":"202610130921da30","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"embedl/bevformer-base","hardware":"l4","variant":"int8","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · one nuScenes sample · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"GPU compute, mean","value":456.95,"unit":"ms","higherIsBetter":false,"digits":2,"baseline":455.13,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":460.75,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":2.19,"unit":"fps","higherIsBetter":true,"digits":2}]},{"title":"Accuracy","subtitle":"nuScenes val · 6,019 samples · PyTorch FP16 with INT8 fake quantization on an NVIDIA L4, not this engine","metrics":[{"description":"NDS (nuScenes val)","value":51.1,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"mAP (nuScenes val)","value":40.9,"unit":"%","higherIsBetter":true,"digits":1}]}],"labels":["Image","Detection"],"build":"20261013092195f3","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"embedl/bevformer-small","upstream":"fundamentalvision/BEVFormer","upstreamUrl":"https://github.com/fundamentalvision/BEVFormer","source":"fundamentalvision/BEVFormer","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=embedl%2Fbevformer-small&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"embedl/bevformer-small","hardware":"orin","variant":"fp16","precision":"FP16","metrics":[{"title":"Latency","subtitle":"batch 1 · one nuScenes sample · GPU compute only","metrics":[{"description":"GPU compute, mean","value":295.2,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"GPU compute, p99","value":295.75,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":3.39,"unit":"fps","higherIsBetter":true,"digits":2}]},{"title":"Accuracy","subtitle":"nuScenes val · 6,019 samples · PyTorch FP16 on an NVIDIA L4, not this engine","metrics":[{"description":"NDS (nuScenes val)","value":47.9,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"mAP (nuScenes val)","value":37.1,"unit":"%","higherIsBetter":true,"digits":1}]}],"labels":["Image","Detection"],"build":"2026101309218a33","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"embedl/bevformer-small","hardware":"orin","variant":"int8","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · one nuScenes sample · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"GPU compute, mean","value":301.49,"unit":"ms","higherIsBetter":false,"digits":2,"baseline":295.2,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":301.77,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":3.32,"unit":"fps","higherIsBetter":true,"digits":2}]},{"title":"Accuracy","subtitle":"nuScenes val · 6,019 samples · PyTorch FP16 with INT8 fake quantization on an NVIDIA L4, not this engine","metrics":[{"description":"NDS (nuScenes val)","value":47.1,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"mAP (nuScenes val)","value":36.5,"unit":"%","higherIsBetter":true,"digits":1}]}],"labels":["Image","Detection"],"build":"2026101309215448","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"embedl/bevformer-small","hardware":"l4","variant":"fp16","precision":"FP16","metrics":[{"title":"Latency","subtitle":"batch 1 · one nuScenes sample · GPU compute only","metrics":[{"description":"GPU compute, mean","value":116.73,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"GPU compute, p99","value":118.61,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":8.57,"unit":"fps","higherIsBetter":true,"digits":2}]},{"title":"Accuracy","subtitle":"nuScenes val · 6,019 samples · PyTorch FP16 on an NVIDIA L4, not this engine","metrics":[{"description":"NDS (nuScenes val)","value":47.9,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"mAP (nuScenes val)","value":37.1,"unit":"%","higherIsBetter":true,"digits":1}]}],"labels":["Image","Detection"],"build":"202610130921f66a","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"embedl/bevformer-small","hardware":"l4","variant":"int8","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · one nuScenes sample · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"GPU compute, mean","value":111.91,"unit":"ms","higherIsBetter":false,"digits":2,"baseline":116.73,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":113.83,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":8.94,"unit":"fps","higherIsBetter":true,"digits":2}]},{"title":"Accuracy","subtitle":"nuScenes val · 6,019 samples · PyTorch FP16 with INT8 fake quantization on an NVIDIA L4, not this engine","metrics":[{"description":"NDS (nuScenes val)","value":47.1,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"mAP (nuScenes val)","value":36.5,"unit":"%","higherIsBetter":true,"digits":1}]}],"labels":["Image","Detection"],"build":"2026101309212d50","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"embedl/bevformer-tiny","upstream":"fundamentalvision/BEVFormer","upstreamUrl":"https://github.com/fundamentalvision/BEVFormer","source":"fundamentalvision/BEVFormer","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=embedl%2Fbevformer-tiny&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"embedl/bevformer-tiny","hardware":"orin","variant":"fp16","precision":"FP16","metrics":[{"title":"Latency","subtitle":"batch 1 · one nuScenes sample · GPU compute only","metrics":[{"description":"GPU compute, mean","value":39.41,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"GPU compute, p99","value":39.52,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":25.37,"unit":"fps","higherIsBetter":true,"digits":2}]},{"title":"Accuracy","subtitle":"nuScenes val · 6,019 samples · PyTorch FP16 on an NVIDIA L4, not this engine","metrics":[{"description":"NDS (nuScenes val)","value":35.4,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"mAP (nuScenes val)","value":25.2,"unit":"%","higherIsBetter":true,"digits":1}]}],"labels":["Image","Detection"],"build":"20261013092199a6","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"embedl/bevformer-tiny","hardware":"orin","variant":"int8","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · one nuScenes sample · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"GPU compute, mean","value":39.72,"unit":"ms","higherIsBetter":false,"digits":2,"baseline":39.41,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":39.76,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":25.17,"unit":"fps","higherIsBetter":true,"digits":2}]},{"title":"Accuracy","subtitle":"nuScenes val · 6,019 samples · PyTorch FP16 with INT8 fake quantization on an NVIDIA L4, not this engine","metrics":[{"description":"NDS (nuScenes val)","value":30.8,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"mAP (nuScenes val)","value":22.2,"unit":"%","higherIsBetter":true,"digits":1}]}],"labels":["Image","Detection"],"build":"2026101309211e36","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"embedl/bevformer-tiny","hardware":"l4","variant":"fp16","precision":"FP16","metrics":[{"title":"Latency","subtitle":"batch 1 · one nuScenes sample · GPU compute only","metrics":[{"description":"GPU compute, mean","value":18.44,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"GPU compute, p99","value":19.27,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":54.22,"unit":"fps","higherIsBetter":true,"digits":2}]},{"title":"Accuracy","subtitle":"nuScenes val · 6,019 samples · PyTorch FP16 on an NVIDIA L4, not this engine","metrics":[{"description":"NDS (nuScenes val)","value":35.4,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"mAP (nuScenes val)","value":25.2,"unit":"%","higherIsBetter":true,"digits":1}]}],"labels":["Image","Detection"],"build":"202610130921cb09","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"embedl/bevformer-tiny","hardware":"l4","variant":"int8","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · one nuScenes sample · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"GPU compute, mean","value":16.27,"unit":"ms","higherIsBetter":false,"digits":2,"baseline":18.44,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":16.75,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":61.46,"unit":"fps","higherIsBetter":true,"digits":2}]},{"title":"Accuracy","subtitle":"nuScenes val · 6,019 samples · PyTorch FP16 with INT8 fake quantization on an NVIDIA L4, not this engine","metrics":[{"description":"NDS (nuScenes val)","value":30.8,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"mAP (nuScenes val)","value":22.2,"unit":"%","higherIsBetter":true,"digits":1}]}],"labels":["Image","Detection"],"build":"2026101309216fb2","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/gcvit_base.in1k","title":"GC ViT Base","source":"timm/gcvit_base.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fgcvit_base.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/gcvit_base.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.57014,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.69836,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":385.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.65,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5a64699224fee84a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcvit_base.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.77423,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.83929,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":172.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.65,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"56452361021cdb9a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcvit_base.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":11.3282,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":11.3496,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":87.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.65,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f0e8917c63620a87","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcvit_base.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.91904,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.99884,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":342.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.65,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.54,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.02,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"d39c3327817cdaae","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcvit_base.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.84664,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.85651,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":259.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.65,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6520b17270b2f930","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/gcvit_small.in1k","title":"GC ViT Small","source":"timm/gcvit_small.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fgcvit_small.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/gcvit_small.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.27571,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.41138,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":433.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.55,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.11,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b402d145d1fb2723","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcvit_small.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.87388,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.88849,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":204.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.55,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.11,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"fd2700a31585cc8d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcvit_small.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.17796,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":9.20837,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":108.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.55,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.11,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7e097071bb83e815","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcvit_small.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.25767,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.3186,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":442.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.55,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.11,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.1,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.58,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"bd4bd1f5ada5c4d2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcvit_small.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.13295,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.14276,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":319.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.55,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.11,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d9c1fc901341e083","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/gcvit_tiny.in1k","title":"GC ViT Tiny","source":"timm/gcvit_tiny.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fgcvit_tiny.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/gcvit_tiny.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.81525,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.96613,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":546.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.09,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b43a6113907ecb41","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcvit_tiny.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.8663,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.87689,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":257.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.09,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b581a943bfd2afa7","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcvit_tiny.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.97741,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":7.00067,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":142.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.09,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"483db47e6666f1cb","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcvit_tiny.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.80033,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.8042,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":555,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.09,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.84,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.32,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"45c8a387cb1c8e6a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcvit_tiny.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.56616,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.58301,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":389.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.09,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"bd2f70217871c0e8","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/gcvit_xxtiny.in1k","title":"GC ViT XXTiny","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fgcvit_xxtiny.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/gcvit_xxtiny.in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":906.006,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":908.63,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":1101.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":15.97,"unit":"%","headline":true},{"description":"Top-5","value":28.86,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/gc_efficientnetv2_rw_t.agc_in1k","title":"GC-EfficientNetV2-RW (t)","source":"timm/gc_efficientnetv2_rw_t.agc_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fgc_efficientnetv2_rw_t.agc_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/gc_efficientnetv2_rw_t.agc_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.84146,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.84741,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":542.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.86,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.34,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.69,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.17,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"7718959fbf404893","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/gcresnet33ts.ra2_in1k","title":"GC-ResNet-33-TS","source":"timm/gcresnet33ts.ra2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fgcresnet33ts.ra2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/gcresnet33ts.ra2_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.720895,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.796143,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1353.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.5,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.76,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ab1f179c8f4860af","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcresnet33ts.ra2_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.88529,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.93677,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":527.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.5,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.76,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"429c9d8859d8174a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcresnet33ts.ra2_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.39033,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.40002,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":293.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.5,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.76,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"15412bdc0905c76d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcresnet33ts.ra2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.40631,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.41101,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":710.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.5,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.76,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.81,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.29,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"a5be9dfe8740d40a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcresnet33ts.ra2_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.69867,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.70898,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":588.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.5,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.76,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1345a5dc7c73a3b6","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/gcresnet50t.ra2_in1k","title":"GC-ResNet-50-T","source":"timm/gcresnet50t.ra2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fgcresnet50t.ra2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/gcresnet50t.ra2_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.784655,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.906187,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.864014,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.99646,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1230.1,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1084.37,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.25,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.05,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ca0b79b65027b79f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcresnet50t.ra2_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.08988,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.10657,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":476.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.25,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.05,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"fb08cc55731d1de4","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcresnet50t.ra2_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.41816,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.42914,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":291,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.25,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.05,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3a03657436e8b25b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcresnet50t.ra2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.732144,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.733398,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1363.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.25,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.05,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.81,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.29,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"f9cfb97e57c5e2e4","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcresnet50t.ra2_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.35334,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.35577,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":738.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.25,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.05,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"733785f8ec42188e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/gcresnext26ts.ch_in1k","title":"GC-ResNeXt-26-TS","source":"timm/gcresnext26ts.ch_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fgcresnext26ts.ch_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/gcresnext26ts.ch_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.504266,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.569275,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1925,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.47,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.63,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0d01417f0b0207cf","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcresnext26ts.ch_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.31049,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.31573,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":758.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.47,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.63,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ad964ca615ec963a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcresnext26ts.ch_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.38371,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.39001,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":417.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.47,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.63,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"cc56511b33dd18a9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcresnext26ts.ch_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.627326,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.631226,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1590.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.47,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.63,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.67,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.15,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"0658b5342565568d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcresnext26ts.ch_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.881862,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.884399,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1132.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.47,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.63,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"51761ebfbbcceb66","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/gcresnext50ts.ch_in1k","title":"GC-ResNeXt-50-TS","source":"timm/gcresnext50ts.ch_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fgcresnext50ts.ch_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/gcresnext50ts.ch_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.986262,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.07764,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":996.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.18,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a9744e4b03553d5d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcresnext50ts.ch_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.34451,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.35553,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":424.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.18,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7e3f44ac8a9e6177","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcresnext50ts.ch_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.28424,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.29456,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":232.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.18,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e2155196a9e03684","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcresnext50ts.ch_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.25904,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.26422,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":793.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.18,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.74,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.22,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"30f3736d2ca74808","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/gcresnext50ts.ch_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.57821,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.58704,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":633.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.18,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"56365e98b58448fb","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"google/gemma-4-12B-it","title":"Gemma 4 12B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=google%2Fgemma-4-12B-it&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"google/gemma-4-12B-it","hardware":"thor","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":39.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":1967,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":58.2,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":86.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":88.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":96.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":95.6,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":88.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":89.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":76.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":79.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.91,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.53,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":75.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":95.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":22.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":14.3,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":11.24,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":10.23,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"fc88783d2b83ec35","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"google/gemma-4-12B-it","hardware":"thor","variant":"huggingface_int4_embedl-flashhead-mtp_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":39.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":1967,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":58.2,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":86.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":88.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":96.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":95.6,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":88.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":89.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":76.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":79.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.91,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.53,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":75.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":95.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":22.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":14.3,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":11.24,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":10.23,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"fc88783d2b83ec35","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"google/gemma-4-12B-it","hardware":"orin","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":24.1,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":1048,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":37.8,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":86.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":88.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":96.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":95.6,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":87.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":89.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":76.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":79.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":2,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.5,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":48.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":58.8,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.4,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":10.76,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":11.12,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":10.2,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"google/gemma-4-12B-it","hardware":"orin","variant":"huggingface_int4_embedl-flashhead-mtp_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":24.1,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":1048,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":37.8,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":86.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":88.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":96.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":95.6,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":87.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":89.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":76.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":79.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":2,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.5,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":48.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":58.8,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.4,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":10.76,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":11.12,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":10.2,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"c09dba6b9a5a2174","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"google/gemma-4-26B-A4B-it","title":"Gemma 4 26B-A4B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=google%2Fgemma-4-26B-A4B-it&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"google/gemma-4-26B-A4B-it","hardware":"thor","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":58.4,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":3861,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":76.4,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":89.3,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"GSM8K","value":96.6,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"IFEval","value":88.2,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"MMLU-Pro","value":83.2,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.2,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.83,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":70.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":78.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":22.9,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":21.54,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":18.84,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":17.83,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"d89a196778b10a14","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"google/gemma-4-26B-A4B-it","hardware":"thor","variant":"huggingface_int4_embedl-mtp_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":58.4,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":3861,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":76.4,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":89.3,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"GSM8K","value":96.6,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"IFEval","value":88.2,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"MMLU-Pro","value":83.2,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.2,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.83,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":70.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":78.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":22.9,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":21.54,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":18.84,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":17.83,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"d89a196778b10a14","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"google/gemma-4-31B-it","title":"Gemma 4 31B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=google%2Fgemma-4-31B-it&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"google/gemma-4-31B-it","hardware":"thor","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":18.1,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":763,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":28.9,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":91.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":91.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":97.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":96.9,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":90.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":91.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":86.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":86.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":4.51,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.22,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":81.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":102.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":25,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":29.16,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":23.2,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":22.19,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"4cf68acacc79f0b8","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"google/gemma-4-31B-it","hardware":"thor","variant":"huggingface_int4_embedl-mtp_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":18.1,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":763,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":28.9,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":91.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":91.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":97.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":96.9,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":90.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":91.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":86.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":86.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":4.51,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.22,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":81.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":102.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":25,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":29.16,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":23.2,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":22.19,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"4cf68acacc79f0b8","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"google/gemma-4-E2B-it","title":"Gemma 4 E2B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=google%2Fgemma-4-E2B-it&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"google/gemma-4-E2B-it","hardware":"thor","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":120,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":46.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":9147,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":16423,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":150.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":48.5,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":74.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":76.6,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":90.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":90.9,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":76.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":78.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":57.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":60.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.55,"unit":"J","higherIsBetter":false,"digits":2,"baseline":1.28,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":1.82,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":0.78,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":66,"unit":"W","higherIsBetter":false,"digits":1,"baseline":59.7,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":74.2,"unit":"W","higherIsBetter":false,"digits":1,"baseline":71.4,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":22.5,"unit":"W","higherIsBetter":false,"digits":1,"baseline":22.8,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":10.03,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":12.35,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":8.38,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":7.37,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"b01f8a285fb11843","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"google/gemma-4-E2B-it","hardware":"thor","variant":"huggingface_fp16_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":46.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":16423,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":48.5,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":76.6,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"GSM8K","value":90.9,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"IFEval","value":78.4,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"MMLU-Pro","value":60.5,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.28,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.78,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":59.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":71.4,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":22.8,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":12.35,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"google/gemma-4-E2B-it","hardware":"thor","variant":"huggingface_int4_embedl-mtp_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":120,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":46.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":9147,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":16423,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":150.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":48.5,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":74.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":76.6,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":90.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":90.9,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":76.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":78.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":57.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":60.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.55,"unit":"J","higherIsBetter":false,"digits":2,"baseline":1.28,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":1.82,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":0.78,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":66,"unit":"W","higherIsBetter":false,"digits":1,"baseline":59.7,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":74.2,"unit":"W","higherIsBetter":false,"digits":1,"baseline":71.4,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":22.5,"unit":"W","higherIsBetter":false,"digits":1,"baseline":22.8,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":10.03,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":12.35,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":8.38,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":7.37,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"b01f8a285fb11843","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"google/gemma-4-E2B-it","hardware":"orin","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":72.9,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":3975,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":102.4,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":74.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":76.6,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":90.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":90.9,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":77.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":78.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":55.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":60.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.53,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.88,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":38.8,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":45.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":11.8,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":7.63,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":8.28,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":7.36,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"google/gemma-4-E2B-it","hardware":"orin","variant":"huggingface_int4_embedl-mtp_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":72.9,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":3975,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":102.4,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":74.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":76.6,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":90.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":90.9,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":77.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":78.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":55.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":60.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.53,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.88,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":38.8,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":45.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":11.8,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":7.63,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":8.28,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":7.36,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"c26c9e3ea58e6b58","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"google/gemma-4-E4B-it","title":"Gemma 4 E4B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=google%2Fgemma-4-E4B-it&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"google/gemma-4-E4B-it","hardware":"thor","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":82.8,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":24.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":5006,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":9679,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":111.8,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":25,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":80.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":82.6,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":92.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":92.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":81.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":86,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":66.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":69.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.84,"unit":"J","higherIsBetter":false,"digits":2,"baseline":2.87,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":1.19,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":0.35,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":69.8,"unit":"W","higherIsBetter":false,"digits":1,"baseline":69.5,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":83.2,"unit":"W","higherIsBetter":false,"digits":1,"baseline":88.6,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":22.2,"unit":"W","higherIsBetter":false,"digits":1,"baseline":22.8,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":13.56,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":18.06,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":11.36,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":10.35,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"a58ffb8a50db7d28","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"google/gemma-4-E4B-it","hardware":"thor","variant":"huggingface_fp16_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":24.2,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":9679,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":25,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":82.6,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"GSM8K","value":92.3,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"IFEval","value":86,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"MMLU-Pro","value":69.5,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":2.87,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.35,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":69.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":88.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":22.8,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":18.06,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"google/gemma-4-E4B-it","hardware":"thor","variant":"huggingface_int4_embedl-flashhead-mtp_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":82.8,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":24.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":5006,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":9679,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":111.8,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":25,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":80.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":82.6,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":92.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":92.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":81.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":86,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":66.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":69.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.84,"unit":"J","higherIsBetter":false,"digits":2,"baseline":2.87,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":1.19,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":0.35,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":69.8,"unit":"W","higherIsBetter":false,"digits":1,"baseline":69.5,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":83.2,"unit":"W","higherIsBetter":false,"digits":1,"baseline":88.6,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":22.2,"unit":"W","higherIsBetter":false,"digits":1,"baseline":22.8,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":13.56,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":18.06,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":11.36,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":10.35,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"a58ffb8a50db7d28","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/ghostnet_100.in1k","title":"GhostNet 1.0×","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fghostnet_100.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/ghostnet_100.in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":593.872,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":594.482,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":1679.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":64.98,"unit":"%","headline":true},{"description":"Top-5","value":86.25,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/ghostnetv2_130.in1k","title":"GhostNetV2 1.3×","source":"timm/ghostnetv2_130.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fghostnetv2_130.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/ghostnetv2_130.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.15431,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.16315,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":461.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.92,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.94,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6604f47e8ee7d14e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/ghostnetv2_130.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.55689,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.60522,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":279.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.92,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.94,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3c8c4e8426624c06","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/ghostnetv2_130.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.760547,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.76178,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1312.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":75.92,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.94,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.65,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.13,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"40e4fa94c886c52a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/ghostnetv2_130.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.0851,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.09595,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":920.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.92,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.94,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e0cad3f40c0d3134","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/ghostnetv2_160.in1k","title":"GhostNetV2 1.6×","source":"timm/ghostnetv2_160.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fghostnetv2_160.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/ghostnetv2_160.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.31139,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.31946,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":430.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.16,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.62,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b7b3f3cc009a5a56","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/ghostnetv2_160.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.97069,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.97968,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":250.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.16,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.62,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"08b4d8a3b6a4e23e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/ghostnetv2_160.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.952809,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.954651,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1047.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.16,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.62,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.7,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.17,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"a4e96a4bf59d0cd2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/ghostnetv2_160.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.17504,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.17786,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":850.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.16,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.62,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3d8b7404288fa62d","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hardcorenas_a.miil_green_in1k","title":"HardCoRe-NAS-A","source":"timm/hardcorenas_a.miil_green_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhardcorenas_a.miil_green_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hardcorenas_a.miil_green_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.370771,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.371765,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2686.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":75.8,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.48,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.07,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"3e61a8548ff0b161","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hardcorenas_b.miil_green_in1k","title":"HardCoRe-NAS-B","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhardcorenas_b.miil_green_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hardcorenas_b.miil_green_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":326.172,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":327.026,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":3051.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":1.54,"unit":"%","headline":true},{"description":"Top-5","value":5.22,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hardcorenas_c.miil_green_in1k","title":"HardCoRe-NAS-C","source":"timm/hardcorenas_c.miil_green_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhardcorenas_c.miil_green_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hardcorenas_c.miil_green_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.474539,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.475525,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2101.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.81,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.08,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.07,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"73971bb546d45c45","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hardcorenas_d.miil_green_in1k","title":"HardCoRe-NAS-D","source":"timm/hardcorenas_d.miil_green_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhardcorenas_d.miil_green_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hardcorenas_d.miil_green_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.629654,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.631287,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1584.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.36,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.43,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.62,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.1,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"fed1e58572af845a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hardcorenas_e.miil_green_in1k","title":"HardCoRe-NAS-E","source":"timm/hardcorenas_e.miil_green_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhardcorenas_e.miil_green_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hardcorenas_e.miil_green_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.620356,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.622162,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1608.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.81,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.6,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.62,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.1,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"0c29cc139f6c1e66","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hardcorenas_f.miil_green_in1k","title":"HardCoRe-NAS-F","source":"timm/hardcorenas_f.miil_green_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhardcorenas_f.miil_green_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hardcorenas_f.miil_green_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.618386,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.620544,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1613.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.93,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.83,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.63,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.1,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"11d4251eb40b2356","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"NousResearch/Hermes-3-Llama-3.1-8B","title":"Hermes 3 Llama 3.1 8B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=NousResearch%2FHermes-3-Llama-3.1-8B&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"NousResearch/Hermes-3-Llama-3.1-8B","hardware":"orin","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":26.1,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":2096,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":32.4,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":57.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":59.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":77.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":80.6,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":56.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":56.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":39.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":41.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.8,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.55,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":47.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":63.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":1.81,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":6.28,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":5.36,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"375c34043742e5eb","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"NousResearch/Hermes-3-Llama-3.1-8B","hardware":"orin","variant":"huggingface_int4_embedl_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":26.1,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":2096,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":32.4,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":57.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":59.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":77.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":80.6,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":56.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":56.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":39.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":41.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.8,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.55,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":47.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":63.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":1.81,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":6.28,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":5.36,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"375c34043742e5eb","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"hexgrad/kokoro-82m","upstream":"hexgrad/Kokoro-82M","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=hexgrad%2FKokoro-82M&granularity=auto","builds":[{"model":"hexgrad/kokoro-82m","hardware":"orin-nx-16gb","precision":"INT8 + FP16/FP32","metrics":[{"title":"Latency","subtitle":"One sentence to WAV · batch 1 · warm · CPU frontend included","metrics":[{"description":"Text to speech, median","value":479.01,"unit":"ms","higherIsBetter":false,"digits":0,"baseline":552.42,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Text to speech, p95","value":546.89,"unit":"ms","higherIsBetter":false,"digits":0,"baseline":638.81,"baselineLabel":"Baseline TensorRT"},{"description":"Real-time factor, median","value":0.0802,"unit":"s/s","higherIsBetter":false,"digits":4,"baseline":0.0928,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"Whisper transcription of generated speech · Baseline TensorRT as baseline","metrics":[{"description":"Word error rate (Whisper)","value":0.806,"unit":"%","higherIsBetter":false,"digits":3,"baseline":0.806,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Log-mel RMSE versus FP32","value":1.438,"unit":"dB","higherIsBetter":false,"digits":3,"baseline":0.845,"baselineLabel":"Baseline TensorRT"},{"description":"WER change versus FP32","value":0,"unit":"pp","higherIsBetter":false,"digits":1}]},{"title":"Memory","metrics":[{"description":"Peak process memory (RSS)","value":1833.4,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":1806.5,"baselineLabel":"Baseline TensorRT","headline":true}]}],"labels":["Text","Speech","Generation"],"build":"2026101815246821","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"hexgrad/kokoro-82m","hardware":"orin-nx-16gb","variant":"huggingface_fp16_tensorrt","precision":"FP16 + FP32","metrics":[{"title":"Latency","subtitle":"One sentence to WAV · batch 1 · warm · CPU frontend included","metrics":[{"description":"Text to speech, median","value":552.42,"unit":"ms","higherIsBetter":false,"digits":0,"headline":true},{"description":"Text to speech, p95","value":638.81,"unit":"ms","higherIsBetter":false,"digits":0},{"description":"Real-time factor, median","value":0.0928,"unit":"s/s","higherIsBetter":false,"digits":4}]},{"title":"Accuracy","subtitle":"Whisper transcription of generated speech · Kokoro v1.0 CPU FP32 as reference","metrics":[{"description":"Word error rate (Whisper)","value":0.806,"unit":"%","higherIsBetter":false,"digits":3,"headline":true},{"description":"Log-mel RMSE versus FP32","value":0.845,"unit":"dB","higherIsBetter":false,"digits":3},{"description":"WER change versus FP32","value":0,"unit":"pp","higherIsBetter":false,"digits":1}]},{"title":"Memory","metrics":[{"description":"Peak process memory (RSS)","value":1806.5,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]}],"labels":["Text","Speech","Generation"],"build":"2026101815242369","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"hexgrad/kokoro-82m","hardware":"orin-nx-16gb","variant":"huggingface_int8_embedl_tensorrt","precision":"INT8 + FP16/FP32","metrics":[{"title":"Latency","subtitle":"One sentence to WAV · batch 1 · warm · CPU frontend included","metrics":[{"description":"Text to speech, median","value":479.01,"unit":"ms","higherIsBetter":false,"digits":0,"baseline":552.42,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Text to speech, p95","value":546.89,"unit":"ms","higherIsBetter":false,"digits":0,"baseline":638.81,"baselineLabel":"Baseline TensorRT"},{"description":"Real-time factor, median","value":0.0802,"unit":"s/s","higherIsBetter":false,"digits":4,"baseline":0.0928,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"Whisper transcription of generated speech · Baseline TensorRT as baseline","metrics":[{"description":"Word error rate (Whisper)","value":0.806,"unit":"%","higherIsBetter":false,"digits":3,"baseline":0.806,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Log-mel RMSE versus FP32","value":1.438,"unit":"dB","higherIsBetter":false,"digits":3,"baseline":0.845,"baselineLabel":"Baseline TensorRT"},{"description":"WER change versus FP32","value":0,"unit":"pp","higherIsBetter":false,"digits":1}]},{"title":"Memory","metrics":[{"description":"Peak process memory (RSS)","value":1833.4,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":1806.5,"baselineLabel":"Baseline TensorRT","headline":true}]}],"labels":["Text","Speech","Generation"],"build":"2026101815246821","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/hgnet_base.ssld_in1k","title":"HGNet Base","source":"timm/hgnet_base.ssld_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhgnet_base.ssld_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hgnet_base.ssld_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.54258,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.89405,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.55377,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.94763,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":647.6,"unit":"qps","digits":1,"baseline":527.528,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.64,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.3,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.78,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"0ddc95fff93a83b1","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hgnet_small.paddle_in1k","title":"HGNet Small","source":"timm/hgnet_small.paddle_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhgnet_small.paddle_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hgnet_small.paddle_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.606282,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.720505,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.608398,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.721741,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1645.3,"unit":"qps","digits":1,"baseline":1385.28,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.39,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.87,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.79,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.27,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"02bf163588e93cd4","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hgnet_tiny.paddle_in1k","title":"HGNet Tiny","source":"timm/hgnet_tiny.paddle_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhgnet_tiny.paddle_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hgnet_tiny.paddle_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.30381,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.31079,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":766.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.07,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.62,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.73,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.21,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e06da9f59b7562bb","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hgnetv2_b0.ssld_stage1_in22k_in1k","title":"HGNetV2-B0","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhgnetv2_b0.ssld_stage1_in22k_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hgnetv2_b0.ssld_stage1_in22k_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":281.921,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":282.654,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":3526.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":52.71,"unit":"%","headline":true},{"description":"Top-5","value":75.45,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hgnetv2_b4.ssld_stage1_in22k_in1k","title":"HGNetV2-B4","source":"timm/hgnetv2_b4.ssld_stage1_in22k_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhgnetv2_b4.ssld_stage1_in22k_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hgnetv2_b4.ssld_stage1_in22k_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.76509,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.77399,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":563.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.55,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.51,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9380864cd65edeed","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hgnetv2_b4.ssld_stage1_in22k_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.87312,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.88287,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":346.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.55,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.51,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1949087c80041686","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hgnetv2_b4.ssld_stage1_in22k_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.618809,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.619537,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1612.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.55,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.51,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.74,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.22,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"7e56ea858fe74401","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hgnetv2_b4.ssld_stage1_in22k_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.89084,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.896667,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1121.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.55,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.51,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c1a0234edfa8847b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hgnetv2_b5.ssld_stage1_in22k_in1k","title":"HGNetV2-B5","source":"timm/hgnetv2_b5.ssld_stage1_in22k_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhgnetv2_b5.ssld_stage1_in22k_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hgnetv2_b5.ssld_stage1_in22k_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.02751,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.03741,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":328.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.21,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.09,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b7f2327179b83f45","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hgnetv2_b5.ssld_stage1_in22k_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.74377,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.75446,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":209.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.21,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.09,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"92aa5b9de28aceee","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hgnetv2_b5.ssld_stage1_in22k_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.15026,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.15369,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":868.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.21,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.09,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.96,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.44,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"c96dfa314179fa99","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hgnetv2_b5.ssld_stage1_in22k_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.62024,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.62476,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":616.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.21,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.09,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1ce7fdf85c62cc58","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hgnetv2_b6.ssld_stage1_in22k_in1k","title":"HGNetV2-B6","source":"timm/hgnetv2_b6.ssld_stage1_in22k_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhgnetv2_b6.ssld_stage1_in22k_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hgnetv2_b6.ssld_stage1_in22k_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.11555,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.17041,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":472.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.12,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.36,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.84,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"591417805bee5505","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hiera_base_224.mae","title":"Hiera Base (224, MAE)","source":"timm/hiera_base_224.mae","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhiera_base_224.mae&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hiera_base_224.mae","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.46808,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.47253,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":680.5,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.1,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.58,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"abb64a21be7c5a75","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hiera_base_plus_224.mae","title":"Hiera Base Plus (224, MAE)","source":"timm/hiera_base_plus_224.mae","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhiera_base_plus_224.mae&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hiera_base_plus_224.mae","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.7668,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.80585,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":565.5,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.31,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.79,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"aa32f08ea036f2e6","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hiera_huge_224.mae","title":"Hiera Huge (224, MAE)","source":"timm/hiera_huge_224.mae","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhiera_huge_224.mae&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hiera_huge_224.mae","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.101,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":9.84906,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":7.19604,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":10.1045,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":140.8,"unit":"qps","digits":1,"baseline":101.516,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":8.02,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":7.5,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"f005c073f823ae10","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hiera_large_224.mae","title":"Hiera Large (224, MAE)","source":"timm/hiera_large_224.mae","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhiera_large_224.mae&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hiera_large_224.mae","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.12812,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.21832,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":242.1,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.95,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.43,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"f4470bed46bad375","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hiera_small_224.mae","title":"Hiera Small (224, MAE)","source":"timm/hiera_small_224.mae","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhiera_small_224.mae&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hiera_small_224.mae","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.00861,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.01831,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":990,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.91,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.39,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"44084f487db89e3a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hiera_tiny_224.mae_in1k_ft_in1k","title":"Hiera Tiny (224, MAE, IN-1k fine-tuned)","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhiera_tiny_224.mae_in1k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hiera_tiny_224.mae_in1k_ft_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":779.602,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":783.63,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":1280,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":24.23,"unit":"%","headline":true},{"description":"Top-5","value":42.84,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hiera_tiny_224.mae","title":"Hiera Tiny (224, MAE)","source":"timm/hiera_tiny_224.mae","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhiera_tiny_224.mae&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hiera_tiny_224.mae","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.779213,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.782959,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1280.7,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.83,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.31,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"be113bead5cfde46","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hrnet_w18.ms_aug_in1k","title":"HRNet-W18","source":"timm/hrnet_w18.ms_aug_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhrnet_w18.ms_aug_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hrnet_w18.ms_aug_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.89917,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.97095,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":518.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.81,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"198089efd96e9dd0","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w18.ms_aug_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.48489,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.49847,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":221.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.81,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"36b2bb7c642a04f2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w18.ms_aug_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.89657,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.9408,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":144.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.81,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"82034463e3af91c8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w18.ms_aug_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.53345,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.71741,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.53693,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.72052,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":651.4,"unit":"qps","digits":1,"baseline":581.805,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.81,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.76,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.24,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"72469ee0d9eaec1a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w18.ms_aug_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.17894,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.18262,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":314.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.81,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"eaffb63b7c9a4025","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hrnet_w18_small.gluon_in1k","title":"HRNet-W18 Small","source":"timm/hrnet_w18_small.gluon_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhrnet_w18_small.gluon_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hrnet_w18_small.gluon_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.509142,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.556458,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1865.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":73.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.84,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2eadecf227095970","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w18_small.gluon_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.61802,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.62378,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":614.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":73.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.84,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"175f7b15f438cfed","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w18_small.gluon_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.60661,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.61481,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":381.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":73.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.84,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"64f2bbd8f97829b3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w18_small.gluon_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.533726,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.534729,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1868,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":73.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.84,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.67,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.15,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"0580a1a8b806eb22","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w18_small.gluon_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.00109,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.00311,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":997.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":73.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.84,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1a3ef23873939826","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hrnet_w18_small_v2.gluon_in1k","title":"HRNet-W18 Small V2","source":"timm/hrnet_w18_small_v2.gluon_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhrnet_w18_small_v2.gluon_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hrnet_w18_small_v2.gluon_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.00318,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.03058,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":968.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.83,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e23449f496b3d08c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w18_small_v2.gluon_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.0001,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.01355,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":331.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.83,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2024a3b8928b0b7c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w18_small_v2.gluon_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.66785,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.68298,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":213.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.83,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6c4426948d2393c2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w18_small_v2.gluon_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.99361,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.995483,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1004.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.83,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.7,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.18,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"50df2ed89598a136","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w18_small_v2.gluon_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.9429,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.94678,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":514.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.83,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e80378a6a68dcf93","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hrnet_w30.ms_in1k","title":"HRNet-W30","source":"timm/hrnet_w30.ms_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhrnet_w30.ms_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hrnet_w30.ms_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.83558,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.90076,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":537.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.7,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.91,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9daeb8b8d5488ac5","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w30.ms_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.04313,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.0564,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":197.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.7,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.91,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"06fc3a548b5ecec9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w30.ms_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.87326,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":7.88696,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":126.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.7,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.91,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4ec9662992f33763","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w30.ms_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.95374,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.95831,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":511.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.7,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.91,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.95,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.42,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"3f643df5e89adceb","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w30.ms_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.64962,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.65991,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":273.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.7,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.91,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"bd7bb21d55ce171f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hrnet_w32.ms_in1k","title":"HRNet-W32","source":"timm/hrnet_w32.ms_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhrnet_w32.ms_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hrnet_w32.ms_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.88765,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.94711,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":521,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.48,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.87,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"bf0cf5e3484489a3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w32.ms_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.98126,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.99744,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":199.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.48,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.87,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"654bfc277cb337f3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w32.ms_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.7966,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":7.81049,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":127.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.48,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.87,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3c9f3a4b687278cd","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w32.ms_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.89581,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.90125,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":527,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.48,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.87,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.46,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"0480ff64a92177b4","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w32.ms_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.5991,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.60663,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":277.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.48,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.87,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2201a96e78819e3f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hrnet_w40.ms_in1k","title":"HRNet-W40","source":"timm/hrnet_w40.ms_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhrnet_w40.ms_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hrnet_w40.ms_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.13591,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.24176,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":467.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.34,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.18,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"576e0fab41a700d2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w40.ms_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.02122,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.0321,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":165.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.34,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.18,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ae38bf78c8701a55","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w40.ms_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":10.4549,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":10.4744,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":95.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.34,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.18,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3e59bbe6eafb0784","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w40.ms_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.11017,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.3229,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":2.11407,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.40448,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":473.5,"unit":"qps","digits":1,"baseline":430.251,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.34,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.18,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.17,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.65,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"c5d88d4a3bd552a5","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w40.ms_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.35085,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.36035,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":229.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.34,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.18,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1ef0b60cdefd36da","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hrnet_w44.ms_in1k","title":"HRNet-W44","source":"timm/hrnet_w44.ms_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhrnet_w44.ms_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hrnet_w44.ms_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.27082,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.396,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":438.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.36,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.25,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ccc9857676c42991","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w44.ms_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.10731,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.13062,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":162.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.36,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.25,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c23c800431fb3a5b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w44.ms_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":10.1881,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":10.212,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":97.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.36,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.25,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6f12d225fa5770c6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w44.ms_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.25463,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.58098,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":2.25769,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.64783,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":443.3,"unit":"qps","digits":1,"baseline":387.251,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.36,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.25,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.27,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.75,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"d3635c321055c8ef","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w44.ms_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.49273,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.5047,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":222.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.36,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.25,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d2e99f76119e4637","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hrnet_w48.ms_in1k","title":"HRNet-W48","source":"timm/hrnet_w48.ms_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhrnet_w48.ms_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hrnet_w48.ms_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.26784,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.38635,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":439.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.18,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5e9e7c1d027839dc","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w48.ms_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.15077,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.17877,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":161.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.18,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7867c976924a1581","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w48.ms_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":10.4056,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":10.429,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":95.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.18,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f2d90d729036489c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w48.ms_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.18451,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.50033,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":2.18842,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.59711,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":457.5,"unit":"qps","digits":1,"baseline":399.722,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.18,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.39,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.87,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"6ddd4a0503ed1546","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w48.ms_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.46255,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.49377,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":224,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.18,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d08c901878bbc0f9","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hrnet_w48_ssld.paddle_in1k","title":"HRNet-W48 (ssld)","source":"timm/hrnet_w48_ssld.paddle_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhrnet_w48_ssld.paddle_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hrnet_w48_ssld.paddle_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.2367,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.34216,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":445.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.2,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.78,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"baa058f32e1368e3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w48_ssld.paddle_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.77804,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.79224,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":172.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.2,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.78,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e3305cf59e168184","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w48_ssld.paddle_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.49327,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":9.52808,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":104.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.2,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.78,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"aeb7b03aa5692579","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w48_ssld.paddle_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.09977,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.53316,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":2.10364,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.60968,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":475.9,"unit":"qps","digits":1,"baseline":394.495,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.2,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.78,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.39,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.87,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"870f188643c7e1cf","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w48_ssld.paddle_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.25458,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.26807,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":235,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.2,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.78,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a8a8b49e1051a03f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/hrnet_w64.ms_in1k","title":"HRNet-W64","source":"timm/hrnet_w64.ms_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fhrnet_w64.ms_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/hrnet_w64.ms_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.13591,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.43904,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":2.27258,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.53082,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":466.6,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":407.596,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.69,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.29,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3cc6ec2d7948403c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w64.ms_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.22144,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.26843,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":159.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.69,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.29,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ffd78af33b44df4c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w64.ms_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":10.275,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":10.365,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":96.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.69,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.29,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e6ba48f7ac35a8a8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w64.ms_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.31663,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.19587,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":2.35626,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.28418,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":431.4,"unit":"qps","digits":1,"baseline":312.764,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.69,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.29,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.96,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.43,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"4c480e5f1f438339","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/hrnet_w64.ms_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.92578,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.94519,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":203,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.69,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.29,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c6573fa046a7abd5","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/iformer_t.in1k","title":"iFormer-T","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fiformer_t.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/iformer_t.in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":523.682,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":524.292,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":1903.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":47.97,"unit":"%","headline":true},{"description":"Top-5","value":72.01,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/inception_v3","title":"Inception V3","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Finception_v3&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/inception_v3","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.218,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.219,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":820.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":68.96,"unit":"%","headline":true},{"description":"Top-5","value":88.24,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/inception_v3","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.218,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.219,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":820.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":68.96,"unit":"%","headline":true},{"description":"Top-5","value":88.24,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/inception_v3.gluon_in1k","title":"Inception v3 (timm)","source":"timm/inception_v3.gluon_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Finception_v3.gluon_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/inception_v3.gluon_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.649145,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.732644,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.692596,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.797058,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1472.9,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1328.98,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.59,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.22,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"93072500105b8351","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/inception_v3.gluon_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.61037,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.61774,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":617.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.59,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.22,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4c2099707241ab72","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/inception_v3.gluon_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.86429,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.87122,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":347.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.59,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.22,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"657a10c77c2b63d9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/inception_v3.gluon_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.609883,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.611908,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1635,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.59,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.22,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.69,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.17,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"4c668f24a7b15dac","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/inception_v3.gluon_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.33797,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.34045,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":746.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.59,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.22,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b720e67df2f138c0","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/inception_v4.tf_in1k","title":"Inception v4","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Finception_v4.tf_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/inception_v4.tf_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":2.512,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":2.554,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":397.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.1,"unit":"%","headline":true},{"description":"Top-5","value":0.52,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/inception_resnet_v2.tf_ens_adv_in1k","title":"Inception-ResNet V2","source":"timm/inception_resnet_v2.tf_ens_adv_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Finception_resnet_v2.tf_ens_adv_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/inception_resnet_v2.tf_ens_adv_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.74649,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.87524,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":566.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.7,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.87,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ef2522d6d38dc6e8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/inception_resnet_v2.tf_ens_adv_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.20837,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.30304,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":236.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.7,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.87,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"691241a30955e404","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/inception_resnet_v2.tf_ens_adv_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.72429,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":8.74091,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":114,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.7,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.87,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7f4ba05ce4ca5d38","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/inception_resnet_v2.tf_ens_adv_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.92076,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.99939,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":255,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.7,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.87,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.3,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.78,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"3ba7997e7efc8c0c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/inception_resnet_v2.tf_ens_adv_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.87948,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.9198,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":170.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.7,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.87,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4084b8fb11cf66a7","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/inception_next_atto.sail_in1k","title":"InceptionNeXt Atto","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Finception_next_atto.sail_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/inception_next_atto.sail_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":474.243,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":474.976,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":2102.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":70.86,"unit":"%","headline":true},{"description":"Top-5","value":89.97,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/inception_next_atto.sail_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.525918,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.527008,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1896,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":70.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.99,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.57,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":49.8,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"24f313b79bd69082","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/inception_next_base.sail_in1k","title":"InceptionNeXt Base","source":"timm/inception_next_base.sail_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Finception_next_base.sail_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/inception_next_base.sail_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.27279,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.35175,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":439.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.73,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.62,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.98,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"1120a809d7292e55","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/inception_next_small.sail_in1k","title":"InceptionNeXt Small","source":"timm/inception_next_small.sail_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Finception_next_small.sail_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/inception_next_small.sail_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.73319,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.73651,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":576.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.48,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.57,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.08,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.56,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"fce5dbfa63a4a4a1","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/inception_next_tiny.sail_in1k","title":"InceptionNeXt Tiny","source":"timm/inception_next_tiny.sail_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Finception_next_tiny.sail_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/inception_next_tiny.sail_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.902759,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.906982,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1106.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.37,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.05,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.84,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.32,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"be03674e36ea8179","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"autotrust/jev-9b","upstream":"autotrust/JEV-9B","title":"JEV 9B","source":"autotrust/jev-9b","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=autotrust%2FJEV-9B&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"autotrust/jev-9b","hardware":"thor","precision":"W8A8","metrics":[{"title":"Latency","metrics":[{"description":"Median time per decision","value":41,"unit":"ms","higherIsBetter":false,"digits":1,"baseline":80.6,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Accuracy","subtitle":"BoolQ, AG News, MNLI, Banking77 · 800 questions","metrics":[{"description":"Decision accuracy","value":85.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":85.25,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Memory","metrics":[{"description":"Memory, model loaded","value":8.41,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":17.73,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":7.47,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":16.71,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]}],"labels":["Text","Classification"],"build":"84fdffc655711bee","updatedAt":"2026-11-10T15:54:34.000Z"},{"model":"autotrust/jev-9b","hardware":"thor","variant":"huggingface_fp16_tensorrt-edgellm","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median time per decision","value":80.6,"unit":"ms","higherIsBetter":false,"digits":1,"headline":true}]},{"title":"Accuracy","subtitle":"BoolQ, AG News, MNLI, Banking77 · 800 questions","metrics":[{"description":"Decision accuracy","value":85.25,"unit":"%","higherIsBetter":true,"digits":2,"headline":true}]},{"title":"Memory","metrics":[{"description":"Memory, model loaded","value":17.73,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":16.71,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true}]}],"labels":["Text","Classification"],"build":"b3cd24fd2ff3ca8f","updatedAt":"2026-11-10T15:54:34.000Z"},{"model":"autotrust/jev-9b","hardware":"thor","variant":"huggingface_int8_embedl_tensorrt-edgellm","precision":"W8A8","metrics":[{"title":"Latency","metrics":[{"description":"Median time per decision","value":41,"unit":"ms","higherIsBetter":false,"digits":1,"baseline":80.6,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Accuracy","subtitle":"BoolQ, AG News, MNLI, Banking77 · 800 questions","metrics":[{"description":"Decision accuracy","value":85.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":85.25,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Memory","metrics":[{"description":"Memory, model loaded","value":8.41,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":17.73,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":7.47,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":16.71,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]}],"labels":["Text","Classification"],"build":"84fdffc655711bee","updatedAt":"2026-11-10T15:54:34.000Z"}]},{"name":"timm/lambda_resnet26t.c1_in1k","title":"Lambda ResNet-26-T","source":"timm/lambda_resnet26t.c1_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flambda_resnet26t.c1_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/lambda_resnet26t.c1_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.422124,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.476379,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2275.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.33,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"21dcbb4969edf3ce","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/lambda_resnet26t.c1_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.23908,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.25269,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":802,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.33,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"bee4364536e8ae0c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/lambda_resnet26t.c1_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.43726,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.45636,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":408,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.33,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d34a56783125d1df","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/lambda_resnet26t.c1_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.314488,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.381056,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.315918,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.382507,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":3164.9,"unit":"qps","digits":1,"baseline":2615.04,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.33,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.65,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.12,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"10eb4393fae0f7ab","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/lambda_resnet26t.c1_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.53982,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.542419,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1848.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.33,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4c580eb090fbb10c","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/lambda_resnet50ts.a1h_in1k","title":"Lambda ResNet-50-TS","source":"timm/lambda_resnet50ts.a1h_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flambda_resnet50ts.a1h_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/lambda_resnet50ts.a1h_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.0405,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.14166,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":938.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.87,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.07,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"da06012b9569539f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/lambda_resnet50ts.a1h_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.65392,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.71649,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":374.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.87,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.07,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0c7fa4fc75550aeb","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/lambda_resnet50ts.a1h_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.38123,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.40735,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":184.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.87,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.07,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c93615d04d5aa8a8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/lambda_resnet50ts.a1h_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.65653,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.6687,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":603.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.87,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.07,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.83,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.31,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"b10bd0d3ec8084b9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/lambda_resnet50ts.a1h_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.0229,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.06439,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":494.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.87,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.07,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"66ddbeaec4085ea6","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/lcnet_050.ra2_in1k","title":"LCNet 0.5×","source":"timm/lcnet_050.ra2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flcnet_050.ra2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/lcnet_050.ra2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.155498,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.155884,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":6377.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":63.1,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":84.34,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.54,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.02,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"65f3bf78b0833b0a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/lcnet_075.ra2_in1k","title":"LCNet 0.75×","source":"timm/lcnet_075.ra2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flcnet_075.ra2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/lcnet_075.ra2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.17549,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.1763,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":5651.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":68.74,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":88.39,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.55,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.03,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"a70486c01f4a4351","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/lcnet_100.ra2_in1k","title":"LCNet 1.0×","source":"timm/lcnet_100.ra2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flcnet_100.ra2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/lcnet_100.ra2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.189359,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.189941,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":5236.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":72.09,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.37,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.56,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.04,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"6a2abd44508a80da","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/lcnetv2_small.paddle_in1k","title":"LCNetV2 Small","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flcnetv2_small.paddle_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/lcnetv2_small.paddle_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":340.271,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":340.942,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":2927.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":62.5,"unit":"%","headline":true},{"description":"Top-5","value":84.52,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/levit_128s.fb_dist_in1k","title":"LeViT (128s)","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flevit_128s.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/levit_128s.fb_dist_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":362.793,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":363.342,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":2746.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":63.15,"unit":"%","headline":true},{"description":"Top-5","value":84.04,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/levit_conv_128s.fb_dist_in1k","title":"LeViT Conv (128s)","source":"timm/levit_conv_128s.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flevit_conv_128s.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/levit_conv_128s.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.431342,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.432068,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2311.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.38,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.81,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.63,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.11,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"eb68b886904c2e4f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/levit_128.fb_dist_in1k","title":"LeViT-128","source":"timm/levit_128.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flevit_128.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/levit_128.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.458472,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.459106,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2173.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":73.98,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.26,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.65,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.13,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"26cc3b4b46814c58","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/levit_conv_128.fb_dist_in1k","title":"LeViT-128 Conv","source":"timm/levit_conv_128.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flevit_conv_128.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/levit_conv_128.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.538012,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.539001,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1852.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":74.03,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.33,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.65,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.13,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"41fed883e3e12d93","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/levit_192.fb_dist_in1k","title":"LeViT-192","source":"timm/levit_192.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flevit_192.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/levit_192.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.508293,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.509399,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1961.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.44,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.5,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.67,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.15,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"a5e50d7502739098","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/levit_conv_192.fb_dist_in1k","title":"LeViT-192 Conv","source":"timm/levit_conv_192.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flevit_conv_192.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/levit_conv_192.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.608044,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.609924,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1641,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.6,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.59,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.67,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.15,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"718f292e3622d5fe","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/levit_256.fb_dist_in1k","title":"LeViT-256","source":"timm/levit_256.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flevit_256.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/levit_256.fb_dist_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.729029,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.815796,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1343.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.72,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"eb3dd4e38d84790f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/levit_256.fb_dist_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.57048,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.58734,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":633,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.72,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f907a00b8b764219","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/levit_256.fb_dist_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.46497,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.48602,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":403.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.72,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"928e685a983c5605","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/levit_256.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.767339,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.769958,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1300.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.72,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.78,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.26,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"812311a49f0ece3f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/levit_256.fb_dist_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.11469,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.11676,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":896.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.72,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ba148c7e8602038c","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/levit_conv_256.fb_dist_in1k","title":"LeViT-256 Conv","source":"timm/levit_conv_256.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flevit_conv_256.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/levit_conv_256.fb_dist_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.740575,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.836853,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1326.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.94,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.77,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"cc54bc2c489e6a0e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/levit_conv_256.fb_dist_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.45383,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.46167,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":683.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.94,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.77,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e6b34a0b4ed7091b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/levit_conv_256.fb_dist_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.43241,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.43951,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":408.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.94,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.77,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"cafe957160a59863","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/levit_conv_256.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.85084,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.853333,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1173.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.94,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.77,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.78,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.26,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"ee8abb97967be94e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/levit_conv_256.fb_dist_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.09201,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.10059,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":914.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.94,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.77,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5c2277fff46976a6","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/levit_384.fb_dist_in1k","title":"LeViT-384","source":"timm/levit_384.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flevit_384.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/levit_384.fb_dist_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.937011,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.03436,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1043.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.03,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.42,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"34b19c58e427f453","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/levit_384.fb_dist_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.98847,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.01099,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":500,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.03,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.42,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1296bee0af354b06","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/levit_384.fb_dist_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.49138,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.51648,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":284.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.03,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.42,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0c1a7b0a7d470ae3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/levit_384.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.14789,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.15039,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":870.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.03,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.42,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.06,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.54,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"c481444b58bde0c5","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/levit_384.fb_dist_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.48798,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.49298,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":671.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.03,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.42,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a3881e7888926928","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/levit_conv_384.fb_dist_in1k","title":"LeViT-384 Conv","source":"timm/levit_conv_384.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flevit_conv_384.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/levit_conv_384.fb_dist_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.03045,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.12433,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":953.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.12,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.45,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6c5201e59f6f4dec","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/levit_conv_384.fb_dist_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.92396,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.9339,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":516.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.12,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.45,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"40c0887b682ebee6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/levit_conv_384.fb_dist_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.53269,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.54248,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":281.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.12,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.45,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"38137eb9f231057e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/levit_conv_384.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.281,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.2887,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":779.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.12,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.45,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.06,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.54,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"d64ab392b4243aea","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/levit_conv_384.fb_dist_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.50194,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.50867,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":665.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.12,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.45,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"be8fb2369b0e6323","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"meta-llama/Llama-3.1-8B-Instruct","title":"Llama 3.1 8B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=meta-llama%2FLlama-3.1-8B-Instruct&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"meta-llama/Llama-3.1-8B-Instruct","hardware":"thor","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":63.6,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":15.6,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":3258,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":6910,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":91.8,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":16.1,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":68.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":70.6,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":86,"unit":"%","higherIsBetter":true,"digits":1,"baseline":86.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":71.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":73,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":46.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":52.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.32,"unit":"J","higherIsBetter":false,"digits":2,"baseline":4.21,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":0.76,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":0.24,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":83.9,"unit":"W","higherIsBetter":false,"digits":1,"baseline":65.8,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":95.2,"unit":"W","higherIsBetter":false,"digits":1,"baseline":98.3,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":23.7,"unit":"W","higherIsBetter":false,"digits":1,"baseline":22.5,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":8.1,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":17.12,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":6.78,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":5.77,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"27390dccd2376a2b","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"meta-llama/Llama-3.1-8B-Instruct","hardware":"thor","variant":"huggingface_fp16_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":15.6,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":6910,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":16.1,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":70.6,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"GSM8K","value":86.7,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"IFEval","value":73,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"MMLU-Pro","value":52.2,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":4.21,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.24,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":65.8,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":98.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":22.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":17.12,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"meta-llama/Llama-3.1-8B-Instruct","hardware":"thor","variant":"huggingface_int4_embedl-eagle3_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":63.6,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":15.6,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":3258,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":6910,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":91.8,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":16.1,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":68.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":70.6,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":86,"unit":"%","higherIsBetter":true,"digits":1,"baseline":86.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":71.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":73,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":46.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":52.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.32,"unit":"J","higherIsBetter":false,"digits":2,"baseline":4.21,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":0.76,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":0.24,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":83.9,"unit":"W","higherIsBetter":false,"digits":1,"baseline":65.8,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":95.2,"unit":"W","higherIsBetter":false,"digits":1,"baseline":98.3,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":23.7,"unit":"W","higherIsBetter":false,"digits":1,"baseline":22.5,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":8.1,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":17.12,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":6.78,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":5.77,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"27390dccd2376a2b","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"meta-llama/Llama-3.1-8B-Instruct","hardware":"orin","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":42.2,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":2024,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":63,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":68.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":70.6,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":86,"unit":"%","higherIsBetter":true,"digits":1,"baseline":86.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":71.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":73,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":46.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":52.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.2,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.83,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":50.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":62.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":3.62,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":6.67,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":5.75,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"f371acca38893ff4","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"meta-llama/Llama-3.1-8B-Instruct","hardware":"orin","variant":"huggingface_int4_embedl-eagle3_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":42.2,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":2024,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":63,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":68.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":70.6,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":86,"unit":"%","higherIsBetter":true,"digits":1,"baseline":86.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":71.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":73,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":46.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":52.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.2,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.83,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":50.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":62.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":3.62,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":6.67,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":5.75,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"f371acca38893ff4","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"meta-llama/Llama-3.2-3B-Instruct","title":"Llama 3.2 3B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=meta-llama%2FLlama-3.2-3B-Instruct&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"meta-llama/Llama-3.2-3B-Instruct","hardware":"thor","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":91.5,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":35.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":7450,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":16215,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":113,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":36.4,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":60.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":61.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":79.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":79.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":66.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":67.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":34,"unit":"%","higherIsBetter":true,"digits":1,"baseline":37.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.75,"unit":"J","higherIsBetter":false,"digits":2,"baseline":1.76,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":1.34,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":0.57,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":68.2,"unit":"W","higherIsBetter":false,"digits":1,"baseline":62.1,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":79.1,"unit":"W","higherIsBetter":false,"digits":1,"baseline":79.2,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":22.6,"unit":"W","higherIsBetter":false,"digits":1,"baseline":24.7,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":4.67,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":8.68,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":3.93,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":2.92,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"fa6968d9848847ef","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"meta-llama/Llama-3.2-3B-Instruct","hardware":"thor","variant":"huggingface_fp16_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":35.4,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":16215,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":36.4,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":61.5,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"GSM8K","value":79.2,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"IFEval","value":67.8,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"MMLU-Pro","value":37.4,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.76,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.57,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":62.1,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":79.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":24.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":8.68,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"meta-llama/Llama-3.2-3B-Instruct","hardware":"thor","variant":"huggingface_int4_embedl-flashhead_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":91.5,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":35.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":7450,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":16215,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":113,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":36.4,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":60.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":61.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":79.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":79.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":66.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":67.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":34,"unit":"%","higherIsBetter":true,"digits":1,"baseline":37.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.75,"unit":"J","higherIsBetter":false,"digits":2,"baseline":1.76,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":1.34,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":0.57,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":68.2,"unit":"W","higherIsBetter":false,"digits":1,"baseline":62.1,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":79.1,"unit":"W","higherIsBetter":false,"digits":1,"baseline":79.2,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":22.6,"unit":"W","higherIsBetter":false,"digits":1,"baseline":24.7,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":4.67,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":8.68,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":3.93,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":2.92,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"fa6968d9848847ef","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"meta-llama/Llama-3.2-3B-Instruct","hardware":"orin","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":58.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":4681,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":72.9,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":59.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":61.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":78.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":79.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":66.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":67.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":34.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":37.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.73,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.36,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":43,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":54.8,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":1,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":3.83,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":2.91,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"4bf7cb50adf9fb26","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"meta-llama/Llama-3.2-3B-Instruct","hardware":"orin","variant":"huggingface_int4_embedl-flashhead_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":58.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":4681,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":72.9,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":59.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":61.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":78.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":79.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":66.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":67.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":34.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":37.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.73,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.36,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":43,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":54.8,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":1,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":3.83,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":2.91,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"4bf7cb50adf9fb26","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"meta-llama/Llama-3.2-3B-Instruct","hardware":"orin-nx-16gb","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":28.9,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":1766,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":38.8,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":60.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":61.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":78.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":79.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":67.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":67.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":35.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":37.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.72,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.38,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":20.9,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":24.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":8.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":2.71,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":3.83,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":2.91,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"meta-llama/Llama-3.2-3B-Instruct","hardware":"orin-nx-16gb","variant":"huggingface_int4_embedl-flashhead_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":28.9,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":1766,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":38.8,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":60.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":61.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":78.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":79.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":67.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":67.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":35.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":37.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.72,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.38,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":20.9,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":24.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":8.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":2.71,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":3.83,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":2.91,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/lowformer_b0.in1k","title":"LowFormer-B0","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flowformer_b0.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/lowformer_b0.in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":631.531,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":632.812,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":1580.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":9.83,"unit":"%","headline":true},{"description":"Top-5","value":23.22,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/lraspp_mobilenet_v3_large","title":"LR-ASPP MobileNetV3 Large","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Flraspp_mobilenet_v3_large&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/lraspp_mobilenet_v3_large","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.521,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.525,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":657.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"COCO segmentation validation · 2,000 images · Baseline TensorRT as baseline","metrics":[{"description":"mIoU","value":43.54,"unit":"%","headline":true}]}],"labels":["Image","Segmentation"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/lraspp_mobilenet_v3_large","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.521,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.525,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":657.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"COCO segmentation validation · 2,000 images · Baseline TensorRT as baseline","metrics":[{"description":"mIoU","value":43.54,"unit":"%","headline":true}]}],"labels":["Image","Segmentation"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mambaout_kobe.in1k","title":"MambaOut (kobe)","source":"timm/mambaout_kobe.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmambaout_kobe.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mambaout_kobe.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.751898,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.847351,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1303.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.62,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"fb287c13fd7563e5","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_kobe.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.60185,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.60962,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":382.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.62,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d807d27feaa6533b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_kobe.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.71138,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.72156,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":211.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.62,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"982a8ec0323f0a93","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_kobe.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.879547,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.88092,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1134.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.62,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.63,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.1,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"04acf75e2a5eb1b6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_kobe.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.16238,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.16571,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":859.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.62,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d6286e8713e01d5a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mambaout_base.in1k","title":"MambaOut Base","source":"timm/mambaout_base.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmambaout_base.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mambaout_base.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.28784,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.11892,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":2.39746,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.22937,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":430.9,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":241.229,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.48,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1605e00fb4e4340d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_base.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":11.1701,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":11.2535,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":89.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.48,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c4ae3e3815214844","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_base.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":23.3968,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":23.5818,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":42.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.48,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1ccad3bfb824a280","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_base.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.94914,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.01428,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":338.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.48,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.48,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.95,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"b4d386fff804d94e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_base.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.0518,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.06903,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":246.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.48,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9cbea295e294add3","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mambaout_base_short_rw.sw_e500_in1k","title":"MambaOut Base (short, rw)","source":"timm/mambaout_base_short_rw.sw_e500_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmambaout_base_short_rw.sw_e500_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mambaout_base_short_rw.sw_e500_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.06115,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.16602,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":480,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.28,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.33,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"82b2c50abcc66fc8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_base_short_rw.sw_e500_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.81566,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.84167,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":146,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.28,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.33,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f408f2555286de77","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_base_short_rw.sw_e500_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":14.5124,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":14.6126,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":68.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.28,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.33,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"53821a9c8effc603","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_base_short_rw.sw_e500_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.80067,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.86237,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":356.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.28,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.33,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.53,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"cfe3ca868bd21c8e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_base_short_rw.sw_e500_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.7427,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.75989,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":267.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.28,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.33,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7535b804ba4ca017","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mambaout_base_tall_rw.sw_e500_in1k","title":"MambaOut Base (tall, rw)","source":"timm/mambaout_base_tall_rw.sw_e500_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmambaout_base_tall_rw.sw_e500_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mambaout_base_tall_rw.sw_e500_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.23219,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.62615,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":2.33508,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.72113,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":443.4,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":376.025,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.38,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.43,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"20a7fcacfb0bcfda","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_base_tall_rw.sw_e500_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.60763,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":7.64325,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":130.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.38,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.43,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"71e53680f4b59e1c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_base_tall_rw.sw_e500_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":15.6676,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":15.7083,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":63.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.38,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.43,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1e80e2db0a1aedf6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_base_tall_rw.sw_e500_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.06579,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.13977,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":326,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.38,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.43,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.5,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.98,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"95ddf04fcbca66e8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_base_tall_rw.sw_e500_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.11096,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.13129,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":243.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.38,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.43,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7b219e13158a8ed5","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mambaout_base_wide_rw.sw_e500_in1k","title":"MambaOut Base (wide, rw)","source":"timm/mambaout_base_wide_rw.sw_e500_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmambaout_base_wide_rw.sw_e500_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mambaout_base_wide_rw.sw_e500_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.24471,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.35535,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":439.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.05,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.29,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4d3144cc91ca2333","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_base_wide_rw.sw_e500_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.76736,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":7.79587,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":128.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.05,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.29,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5186268fe9e965e5","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_base_wide_rw.sw_e500_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":16.0696,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":16.0953,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":61.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.05,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.29,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7e4b4f2f743232b2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_base_wide_rw.sw_e500_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.247,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.34058,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":307.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.05,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.29,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.06,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e1a5299f5181f15e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_base_wide_rw.sw_e500_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.55759,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.5766,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":219.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.05,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.29,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8c2b969e21dd40e5","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mambaout_base_plus_rw.sw_e150_in12k_ft_in1k","title":"MambaOut Base Plus (rw)","source":"timm/mambaout_base_plus_rw.sw_e150_in12k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmambaout_base_plus_rw.sw_e150_in12k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mambaout_base_plus_rw.sw_e150_in12k_ft_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.40729,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.49786,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":411.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.77,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"943e41bf74522da9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_base_plus_rw.sw_e150_in12k_ft_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.33947,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":8.37512,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":119.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.77,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"53bdcf2d17464ca6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_base_plus_rw.sw_e150_in12k_ft_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":17.2428,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":17.3025,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":57.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.77,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"de4c91e74e17937c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_base_plus_rw.sw_e150_in12k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.49437,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.55536,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":286,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":85.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.77,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.66,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.14,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"50c0950291421b0e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_base_plus_rw.sw_e150_in12k_ft_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.87471,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.89709,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":205.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.77,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6c8fd2596fda9a4f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mambaout_femto.in1k","title":"MambaOut Femto","source":"timm/mambaout_femto.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmambaout_femto.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mambaout_femto.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.607843,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.696228,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1605.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.09,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.93,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2df3d7a9bb4a402b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_femto.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.10197,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.11102,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":473,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.09,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.93,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"97ed124c2e44121e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_femto.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.84465,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.85504,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":258.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.09,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.93,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e8537323ff852544","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_femto.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.695235,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.696655,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1435.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.09,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.93,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.6,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":85.3,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"93fb02b42996f8b1","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_femto.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.954658,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.958496,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1046.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.09,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.93,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ba2432e5b6aa4915","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mambaout_small.in1k","title":"MambaOut Small","source":"timm/mambaout_small.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmambaout_small.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mambaout_small.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.71648,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.83325,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":575.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.49,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"277e47093f4068e3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_small.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.52514,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":7.65271,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":132.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.49,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5e1cf89bbcc200fc","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_small.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":15.7389,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":15.8303,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":63.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.49,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"aa1dfb83dabf79ea","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_small.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.99981,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.00403,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":499.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.49,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.07,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.55,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"8f7083a870ecf2d6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_small.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.82963,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.84467,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":353.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.49,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b6e4c72779aecb25","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mambaout_small_rw.sw_e450_in1k","title":"MambaOut Small (rw)","source":"timm/mambaout_small_rw.sw_e450_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmambaout_small_rw.sw_e450_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mambaout_small_rw.sw_e450_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.77617,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.90082,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":554.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.94,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.25,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"751be869f3f8ee8d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_small_rw.sw_e450_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.40482,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.42712,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":184.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.94,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.25,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"175e40c194ec3296","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_small_rw.sw_e450_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":10.9369,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":10.9669,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":91,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.94,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.25,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"dad673979abb0fda","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_small_rw.sw_e450_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.0188,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.09027,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":495,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.94,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.25,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.07,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.55,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"ace1369ccac790ee","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_small_rw.sw_e450_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.77223,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.78906,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":360.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.94,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.25,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"57f9880baa35d4ec","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mambaout_tiny.in1k","title":"MambaOut Tiny","source":"timm/mambaout_tiny.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmambaout_tiny.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mambaout_tiny.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.904895,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.01117,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1075.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.56,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3dd4da4cee226aeb","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_tiny.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.40545,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.44135,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":225.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.56,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"11ca4c0b07b5f869","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_tiny.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.2914,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":9.35571,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":107.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.56,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"bf3f7afe84955d5d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_tiny.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.17628,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.17932,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":849.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.56,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.82,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.3,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"f38724cb083c3c32","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mambaout_tiny.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.6413,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.64868,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":608.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.56,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7c962d8ed32c8dbe","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/maxvit_rmlp_pico_rw_256.sw_in1k","title":"MaxViT Pico (256, rmlp, rw)","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmaxvit_rmlp_pico_rw_256.sw_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/maxvit_rmlp_pico_rw_256.sw_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.121,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.123,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":890.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":27.05,"unit":"%","headline":true},{"description":"Top-5","value":47.08,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/maxvit_rmlp_small_rw_224.sw_in1k","title":"MaxViT Small (224, rmlp, rw)","source":"timm/maxvit_rmlp_small_rw_224.sw_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmaxvit_rmlp_small_rw_224.sw_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/maxvit_rmlp_small_rw_224.sw_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.24985,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.3114,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":444.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.69,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.91,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.29,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.77,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"ac72a66545ae3bdf","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/maxvit_xlarge_tf_224.in21k","title":"MaxViT XLarge (224, tf)","source":"timm/maxvit_xlarge_tf_224.in21k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmaxvit_xlarge_tf_224.in21k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/maxvit_xlarge_tf_224.in21k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.82019,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":9.95166,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":101.8,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":5.84,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":5.32,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"854db0a4ffb1ce1e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/maxxvit_rmlp_nano_rw_256.sw_in1k","title":"MaxxViT Nano (256, rmlp, rw)","source":"timm/maxxvit_rmlp_nano_rw_256.sw_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmaxxvit_rmlp_nano_rw_256.sw_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/maxxvit_rmlp_nano_rw_256.sw_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.3889,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.50677,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":710.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.71,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.26,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7ab3118917feeb02","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/maxxvit_rmlp_nano_rw_256.sw_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.85269,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":7.94189,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":126.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.71,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.26,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b3df7b1f99952092","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/maxxvit_rmlp_nano_rw_256.sw_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":12.1847,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":12.2754,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":81.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.71,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.26,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"124be9058c5a569c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/maxxvit_rmlp_nano_rw_256.sw_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.69241,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.69513,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":371.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.71,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.26,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.71,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.19,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"083df8b9b985eab5","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/maxxvit_rmlp_nano_rw_256.sw_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.87336,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.87939,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":205.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.71,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.26,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b874628e6c711978","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/maxxvit_rmlp_small_rw_256.sw_in1k","title":"MaxxViT Small (256, rmlp, rw)","source":"timm/maxxvit_rmlp_small_rw_256.sw_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmaxxvit_rmlp_small_rw_256.sw_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/maxxvit_rmlp_small_rw_256.sw_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.64569,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.91643,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":2.73846,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.03925,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":373.6,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":337.853,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.38,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.96,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"652fdbac068f88db","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/maxxvit_rmlp_small_rw_256.sw_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":13.0057,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":13.1028,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":76.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.38,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.96,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5a5d753bd8847521","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/maxxvit_rmlp_small_rw_256.sw_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":21.714,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":21.8662,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":46,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.38,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.96,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7c24c54fa6ea33c0","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/maxxvit_rmlp_small_rw_256.sw_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.96475,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.96857,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":201.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.38,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.96,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.27,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.74,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"bea04ff2e44158b1","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/maxxvit_rmlp_small_rw_256.sw_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.00348,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":8.01038,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":124.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.38,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.96,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d95e5a42117ef128","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/maxxvitv2_rmlp_base_rw_224.sw_in12k_ft_in1k","title":"MaxxViTv2 Base (224, rmlp, rw)","source":"timm/maxxvitv2_rmlp_base_rw_224.sw_in12k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmaxxvitv2_rmlp_base_rw_224.sw_in12k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/maxxvitv2_rmlp_base_rw_224.sw_in12k_ft_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.49117,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.07646,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":3.60071,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.18518,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":283.5,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":243.777,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":86.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.96,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"fb4b4680d26ae9bb","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/maxxvitv2_rmlp_base_rw_224.sw_in12k_ft_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":16.828,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":17.0875,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":59.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":86.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.96,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1c9f982a5b844f40","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/maxxvitv2_rmlp_base_rw_224.sw_in12k_ft_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":28.1379,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":28.3484,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":35.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":86.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.96,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a6f4a59029448f34","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/maxxvitv2_rmlp_base_rw_224.sw_in12k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.89057,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.89789,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":145.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":86.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.96,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.83,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.31,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"4386044e12ff44d7","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/maxxvitv2_rmlp_base_rw_224.sw_in12k_ft_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":10.9663,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":10.9716,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":91.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":86.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.96,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c2d43fc3b75104f2","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/maxxvitv2_rmlp_base_rw_384.sw_in12k_ft_in1k","title":"MaxxViTv2 Base (384, rmlp, rw)","source":"timm/maxxvitv2_rmlp_base_rw_384.sw_in12k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmaxxvitv2_rmlp_base_rw_384.sw_in12k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/maxxvitv2_rmlp_base_rw_384.sw_in12k_ft_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.82736,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.9502,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":145.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":87.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.27,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0548459690bfaf15","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/maxxvitv2_rmlp_base_rw_384.sw_in12k_ft_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":24.4739,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":24.6189,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":40.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":87.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.27,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0b1b6acfa235b01a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/maxxvitv2_rmlp_base_rw_384.sw_in12k_ft_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":49.7072,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":49.9268,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":20.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":87.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.27,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9969dabad68823cc","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/maxxvitv2_rmlp_base_rw_384.sw_in12k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.93605,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":9.23291,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":111.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":87.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.27,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.83,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.31,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"36d6b14966388fec","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/maxxvitv2_rmlp_base_rw_384.sw_in12k_ft_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":13.8267,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":13.8926,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":72.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":87.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.27,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7959c003745a83f5","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/maxxvitv2_nano_rw_256.sw_in1k","title":"MaxxViTv2 Nano (256, rw)","source":"timm/maxxvitv2_nano_rw_256.sw_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmaxxvitv2_nano_rw_256.sw_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/maxxvitv2_nano_rw_256.sw_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.36188,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.44501,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":717.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.19,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4dea92cbbc383fb0","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/maxxvitv2_nano_rw_256.sw_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.91117,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":8.94171,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":111.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.19,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b6f2b36e59d470cb","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/maxxvitv2_nano_rw_256.sw_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":13.8231,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":13.8807,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":72.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.19,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f5fb0ee2aa4eadef","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/maxxvitv2_nano_rw_256.sw_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.62837,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.68713,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":275.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.19,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.79,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.27,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"308e31818e643b5b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/maxxvitv2_nano_rw_256.sw_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.06036,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.06909,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":165,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.19,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a7be8cd96861cec2","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mixnet_l.ft_in1k","title":"MixNet-L","source":"timm/mixnet_l.ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmixnet_l.ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mixnet_l.ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.68588,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.69055,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":592.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.5,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.04,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.62,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.1,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"34613e3f98940d0a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mixnet_m.ft_in1k","title":"MixNet-M","source":"timm/mixnet_m.ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmixnet_m.ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mixnet_m.ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.08421,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.08606,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":921.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.96,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.29,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.07,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"373cd1be1d706238","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tf_mixnet_m.in1k","title":"MixNet-M (TF)","source":"timm/tf_mixnet_m.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftf_mixnet_m.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tf_mixnet_m.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.58811,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.591,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":629.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.12,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.72,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.06,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"8fc3bb22e9a55a07","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mixnet_s.ft_in1k","title":"MixNet-S","source":"timm/mixnet_s.ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmixnet_s.ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mixnet_s.ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.844968,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.846558,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1181.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":75.74,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.55,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.58,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.05,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"40edde12413dff59","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tf_mixnet_s.in1k","title":"MixNet-S (TF)","source":"timm/tf_mixnet_s.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftf_mixnet_s.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tf_mixnet_s.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.833888,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.835999,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1197.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":75.49,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.5,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.58,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.05,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"84b783ed3bd1a9dd","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mixnet_xl.ra_in1k","title":"MixNet-XL","source":"timm/mixnet_xl.ra_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmixnet_xl.ra_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mixnet_xl.ra_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.31077,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.3139,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":432.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.53,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.68,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.15,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"fa052f0402856b53","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/mnasnet0_5","title":"MNASNet 0.5","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fmnasnet0_5&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/mnasnet0_5","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":292.419,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":292.786,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":3402.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":46.65,"unit":"%","headline":true},{"description":"Top-5","value":70.29,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/mnasnet0_5","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":292.419,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":292.786,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":3402.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":46.65,"unit":"%","headline":true},{"description":"Top-5","value":70.29,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/mnasnet0_75","title":"MNASNet 0.75","source":"torchvision/mnasnet0_75","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fmnasnet0_75&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/mnasnet0_75","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.3789002398843877,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":2639.2171203299476,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":68.786,"unit":"%","digits":2,"baseline":71.18,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":88.856,"unit":"%","digits":2,"baseline":90.496,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"torchvision/mnasnet1_0","title":"MNASNet 1.0","source":"torchvision/mnasnet1_0","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fmnasnet1_0&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/mnasnet1_0","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.3171572368592024,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":3153.010191105733,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":70.65,"unit":"%","digits":2,"baseline":73.456,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":89.744,"unit":"%","digits":2,"baseline":91.51,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/mnasnet_100.rmsp_in1k","title":"MnasNet 1.0×","source":"timm/mnasnet_100.rmsp_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmnasnet_100.rmsp_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mnasnet_100.rmsp_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.158445,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.181767,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.158844,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.182465,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":6248.7,"unit":"qps","digits":1,"baseline":5451.91,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":70.91,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.72,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.57,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":51.7,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"23847333c26a1f54","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/mnasnet1_3","title":"MNASNet 1.3","source":"torchvision/mnasnet1_3","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fmnasnet1_3&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/mnasnet1_3","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.33116879872977734,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":3019.6081389175993,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":74.444,"unit":"%","digits":2,"baseline":76.506,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":92.38,"unit":"%","digits":2,"baseline":93.522,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/mobilenetv1_100h.ra4_e3600_r224_in1k","title":"MobileNetV1 (100h)","source":"timm/mobilenetv1_100h.ra4_e3600_r224_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilenetv1_100h.ra4_e3600_r224_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilenetv1_100h.ra4_e3600_r224_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.0848958,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.113418,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.0853271,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.114288,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":11566.8,"unit":"qps","digits":1,"baseline":8703.64,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":74.02,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.53,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.58,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":61.4,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"8b97bcacd5d458f3","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobilenetv1_100.ra4_e3600_r224_in1k","title":"MobileNetV1 1.0×","source":"timm/mobilenetv1_100.ra4_e3600_r224_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilenetv1_100.ra4_e3600_r224_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilenetv1_100.ra4_e3600_r224_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.0858109,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.108624,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.0861206,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.109802,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":11468.9,"unit":"qps","digits":1,"baseline":9079.01,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":73.22,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.24,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.57,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":49.2,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"2d5d5c1b45874c70","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobilenetv1_125.ra4_e3600_r224_in1k","title":"MobileNetV1 1.25×","source":"timm/mobilenetv1_125.ra4_e3600_r224_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilenetv1_125.ra4_e3600_r224_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilenetv1_125.ra4_e3600_r224_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.100056,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.135366,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.100586,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.136414,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":9842.6,"unit":"qps","digits":1,"baseline":7311.59,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":75.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.48,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":72.6,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"c718683265eb1cd6","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/mobilenet_v2","title":"MobileNetV2","source":"torchvision/mobilenet_v2","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilenetv2_100.ra_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/mobilenet_v2","hardware":"orin-nano-super","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.38037,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":723.543,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNette validation · 3,925 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":74.54777070063695,"unit":"%","digits":2,"baseline":75.46496815286623,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":93.17197452229298,"unit":"%","digits":2,"baseline":93.68152866242038,"baselineLabel":"FP32 model"}]},{"title":"Size","metrics":[{"description":"Engine size","value":4,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"torchvision/mobilenet_v2","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.42283314280211926,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":2364.999094850963,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":71.372,"unit":"%","digits":2,"baseline":72.154,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":90.422,"unit":"%","digits":2,"baseline":90.822,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/mobilenetv2_110d.ra_in1k","title":"MobileNetV2 (110d)","source":"timm/mobilenetv2_110d.ra_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilenetv2_110d.ra_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilenetv2_110d.ra_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.352447,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.42572,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2767.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.74,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.98,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"558fe9d9ee2d3052","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mobilenetv2_110d.ra_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.623685,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.627869,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1591.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.74,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.98,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5f72989ffdb7ffd9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mobilenetv2_110d.ra_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.963601,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.96814,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1030.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.74,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.98,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7cb5d9668f739469","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mobilenetv2_110d.ra_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.185667,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.207096,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.186035,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.209106,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":5342.4,"unit":"qps","digits":1,"baseline":4788.7,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":74.74,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.98,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.57,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":53.5,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"7e2da6177568ae00","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mobilenetv2_110d.ra_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.328025,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.329651,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3038.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.74,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.98,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"23cdc40d23f6f5e5","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobilenetv2_120d.ra_in1k","title":"MobileNetV2 (120d)","source":"timm/mobilenetv2_120d.ra_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilenetv2_120d.ra_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilenetv2_120d.ra_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.417561,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.489746,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2336.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.35,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"60000cc59e297167","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mobilenetv2_120d.ra_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.737336,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.742859,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1346.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.35,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f84a0941650e4ecf","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mobilenetv2_120d.ra_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.19399,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.19836,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":832.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.35,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d9a54fc3171419a1","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mobilenetv2_120d.ra_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.220486,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.254342,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.221436,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.255432,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":4499.5,"unit":"qps","digits":1,"baseline":3906.21,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.35,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":69,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"7ec12f108fcfc337","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mobilenetv2_120d.ra_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.414433,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.415833,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2406.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.35,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5bac24efeb0f38b0","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobilenetv2_050.lamb_in1k","title":"MobileNetV2 0.5×","source":"timm/mobilenetv2_050.lamb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilenetv2_050.lamb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilenetv2_050.lamb_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.217057,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.274902,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":4534.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":64.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":85.04,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ce38508ad0ab585a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mobilenetv2_050.lamb_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.413079,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.417389,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2400.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":64.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":85.04,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"cf675d90a6935e53","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mobilenetv2_050.lamb_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.631372,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.63562,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1571.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":64.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":85.04,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"cb7c3ce742df4360","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mobilenetv2_050.lamb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.131332,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.131653,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":7523.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":64.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":85.04,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.54,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":23.5,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"6d3f2d35683e69a2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mobilenetv2_050.lamb_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.168924,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.169983,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":5884.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":64.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":85.04,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d4cd401260acb0e2","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobilenetv2_100.ra_in1k","title":"MobileNetV2 1.0×","source":"timm/mobilenetv2_100.ra_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilenetv2_100.ra_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilenetv2_100.ra_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.252709,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.310608,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3860,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.69,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.89,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1dca87178ba378d4","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mobilenetv2_100.ra_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.489061,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.492615,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2028.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.69,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.89,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"391b9f7bf747bc34","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mobilenetv2_100.ra_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.738696,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.743042,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1343.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.69,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.89,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b43271bf83eb9731","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mobilenetv2_100.ra_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.142838,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.159399,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.143188,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.161224,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":6936.7,"unit":"qps","digits":1,"baseline":6220.1,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":72.69,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.89,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.56,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":41.5,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"f272192671599bb1","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mobilenetv2_100.ra_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.247214,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.24884,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":4027.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.69,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.89,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0c5041f5c463db95","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobilenetv2_140.ra_in1k","title":"MobileNetV2 1.4×","source":"timm/mobilenetv2_140.ra_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilenetv2_140.ra_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilenetv2_140.ra_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.330078,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.390259,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2912.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.68,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e714c561ed29fbea","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mobilenetv2_140.ra_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.551864,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.555847,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1798.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.68,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f8848db89b36c34c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mobilenetv2_140.ra_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.92155,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.926331,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1077.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.68,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"05d75331fc033abc","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mobilenetv2_140.ra_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.158869,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.192593,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.159912,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.193787,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":6232,"unit":"qps","digits":1,"baseline":5156.3,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":75.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.68,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":71.7,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"76a06e9077a53ed5","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mobilenetv2_140.ra_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.284118,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.285767,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3506.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.68,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"542d20a35aa8e82e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobilenetv3_rw.rmsp_in1k","title":"MobileNetV3 (rw)","source":"timm/mobilenetv3_rw.rmsp_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilenetv3_rw.rmsp_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilenetv3_rw.rmsp_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.43418,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.435425,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2295.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":75.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.66,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.07,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e90b6caba3dbe616","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/mobilenet_v3_large","title":"MobileNetV3 Large","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fmobilenet_v3_large&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/mobilenet_v3_large","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":555.115,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":555.908,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":1797.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":18.76,"unit":"%","headline":true},{"description":"Top-5","value":37.73,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/mobilenet_v3_large","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":555.115,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":555.908,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":1797.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":18.76,"unit":"%","headline":true},{"description":"Top-5","value":37.73,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobilenetv3_large_150d.ra4_e3600_r256_in1k","title":"MobileNetV3 Large (150d)","source":"timm/mobilenetv3_large_150d.ra4_e3600_r256_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilenetv3_large_150d.ra4_e3600_r256_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilenetv3_large_150d.ra4_e3600_r256_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.832584,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.840942,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1199,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.48,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.21,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.71,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.19,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"14fb2a0a3c506f79","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tf_mobilenetv3_large_075.in1k","title":"MobileNetV3 Large 0.75× (TF)","source":"timm/tf_mobilenetv3_large_075.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftf_mobilenetv3_large_075.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tf_mobilenetv3_large_075.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.404015,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.404907,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2464.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":73.25,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.26,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.57,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.05,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"3b341e44463af7f4","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tf_mobilenetv3_large_minimal_100.in1k","title":"MobileNetV3 Large 1.0× (minimal, TF)","source":"timm/tf_mobilenetv3_large_minimal_100.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftf_mobilenetv3_large_minimal_100.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tf_mobilenetv3_large_minimal_100.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.294834,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.295837,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3373.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":72.06,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.52,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.57,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.05,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"727816a0d4dc5492","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tf_mobilenetv3_large_100.in1k","title":"MobileNetV3 Large 1.0× (TF)","source":"timm/tf_mobilenetv3_large_100.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftf_mobilenetv3_large_100.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tf_mobilenetv3_large_100.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.439598,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.44104,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2266.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":75.3,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.38,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.07,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"a119ac00361816a4","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/mobilenet_v3_small","title":"MobileNetV3 Small","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fmobilenet_v3_small&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/mobilenet_v3_small","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":402.222,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":402.71,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":2477.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.29,"unit":"%","headline":true},{"description":"Top-5","value":1.17,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/mobilenet_v3_small","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":402.222,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":402.71,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":2477.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.29,"unit":"%","headline":true},{"description":"Top-5","value":1.17,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobilenetv3_small_050.lamb_in1k","title":"MobileNetV3 Small 0.5×","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilenetv3_small_050.lamb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilenetv3_small_050.lamb_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":256.287,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":256.653,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":3879.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.13,"unit":"%","headline":true},{"description":"Top-5","value":0.55,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobilenetv4_conv_large.e500_r256_in1k","title":"MobileNetV4 Conv Large","source":"timm/mobilenetv4_conv_large.e500_r256_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilenetv4_conv_large.e500_r256_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilenetv4_conv_large.e500_r256_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.361524,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.617195,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.362152,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.618652,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":2753.4,"unit":"qps","digits":1,"baseline":1615.88,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.65,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.15,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.89,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.37,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"4ce36667d50ce7c5","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobilenetv4_conv_aa_large.e230_r384_in12k_ft_in1k","title":"MobileNetV4 Conv Large (aa)","source":"timm/mobilenetv4_conv_aa_large.e230_r384_in12k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilenetv4_conv_aa_large.e230_r384_in12k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilenetv4_conv_aa_large.e230_r384_in12k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.885512,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.888916,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1127.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.81,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.03,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.89,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.37,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"8f55b46563547e74","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobilenetv4_conv_medium.e250_r384_in12k_ft_in1k","title":"MobileNetV4 Conv Medium","source":"timm/mobilenetv4_conv_medium.e250_r384_in12k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilenetv4_conv_medium.e250_r384_in12k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilenetv4_conv_medium.e250_r384_in12k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.772138,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.777222,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1292.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.33,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.64,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.12,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e6c94f5720e0ff57","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobilenetv4_conv_blur_medium.e500_r224_in1k","title":"MobileNetV4 Conv Medium (blur)","source":"timm/mobilenetv4_conv_blur_medium.e500_r224_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilenetv4_conv_blur_medium.e500_r224_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilenetv4_conv_blur_medium.e500_r224_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.392284,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.393677,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2538.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.17,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.73,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.63,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.11,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"7ded22031f9a5c8d","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobilenetv4_conv_small.e1200_r224_in1k","title":"MobileNetV4 Conv Small","source":"timm/mobilenetv4_conv_small.e1200_r224_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilenetv4_conv_small.e1200_r224_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilenetv4_conv_small.e1200_r224_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.129173,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.130493,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":7577.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":70.11,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.45,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.56,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":43.6,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"344aae7080fa5bf9","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobilenetv4_hybrid_large.e600_r384_in1k","title":"MobileNetV4 Hybrid Large","source":"timm/mobilenetv4_hybrid_large.e600_r384_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilenetv4_hybrid_large.e600_r384_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilenetv4_hybrid_large.e600_r384_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.27788,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.3045,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":438.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.7,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.72,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.47,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"7deeb432c9ae811c","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobilenetv4_hybrid_medium.e200_r256_in12k_ft_in1k","title":"MobileNetV4 Hybrid Medium","source":"timm/mobilenetv4_hybrid_medium.e200_r256_in12k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilenetv4_hybrid_medium.e200_r256_in12k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilenetv4_hybrid_medium.e200_r256_in12k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.841259,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.842651,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1186.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.02,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.16,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.66,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.14,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"c0d9d59d8f832acf","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobilenetv5_300m.gemma3n","title":"MobileNetV5 300M","source":"timm/mobilenetv5_300m.gemma3n","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilenetv5_300m.gemma3n&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilenetv5_300m.gemma3n","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":106.024,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":108.76,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":9.4,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.24,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.71,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"0379515852935959","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobileone_s0.apple_in1k","title":"MobileOne-S0","source":"timm/mobileone_s0.apple_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobileone_s0.apple_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobileone_s0.apple_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.6996,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.774999,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.70166,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.776611,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1426.3,"unit":"qps","digits":1,"baseline":1287.76,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":71.23,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.58,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.07,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"3cfc6b1a5896a3cd","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobileone_s1.apple_in1k","title":"MobileOne-S1","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobileone_s1.apple_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobileone_s1.apple_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.058,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.059,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":944.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.38,"unit":"%","headline":true},{"description":"Top-5","value":1.54,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobileone_s2.apple_in1k","title":"MobileOne-S2","source":"timm/mobileone_s2.apple_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobileone_s2.apple_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobileone_s2.apple_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.646905,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.648621,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1542.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.45,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.63,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.1,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"6c17f7d1e145abc4","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobileone_s4.apple_in1k","title":"MobileOne-S4","source":"timm/mobileone_s4.apple_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobileone_s4.apple_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobileone_s4.apple_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.27046,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.27515,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":786.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.07,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.17,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.72,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.2,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"685cadab0d48d0a7","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobilevit_s.cvnets_in1k","title":"MobileViT-S","source":"timm/mobilevit_s.cvnets_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilevit_s.cvnets_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilevit_s.cvnets_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.929072,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.933472,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1074.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.67,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.86,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.6,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.08,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"b52b1968942096d2","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobilevit_xs.cvnets_in1k","title":"MobileViT-XS","source":"timm/mobilevit_xs.cvnets_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilevit_xs.cvnets_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilevit_xs.cvnets_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.76054,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.766632,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1312.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":74.17,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.01,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.55,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.03,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"fe8734665d3f8580","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobilevit_xxs.cvnets_in1k","title":"MobileViT-XXS","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilevit_xxs.cvnets_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilevit_xxs.cvnets_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":402.466,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":403.137,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":2474.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.1,"unit":"%","headline":true},{"description":"Top-5","value":0.48,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobilevitv2_050.cvnets_in1k","title":"MobileViTV2 0.5×","source":"timm/mobilevitv2_050.cvnets_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilevitv2_050.cvnets_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilevitv2_050.cvnets_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.70154,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.703186,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1422.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":69.08,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.31,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.54,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.02,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"86db882a311e55fe","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobilevitv2_100.cvnets_in1k","title":"MobileViTV2 1.0×","source":"timm/mobilevitv2_100.cvnets_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilevitv2_100.cvnets_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilevitv2_100.cvnets_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.942979,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.947205,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1059,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.4,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.79,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.58,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.06,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"8d5b3c3ed9b34275","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobilevitv2_125.cvnets_in1k","title":"MobileViTV2 1.25×","source":"timm/mobilevitv2_125.cvnets_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilevitv2_125.cvnets_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilevitv2_125.cvnets_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.1297,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.13202,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":884,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.34,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.7,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.61,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.09,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"508859511c124ee9","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobilevitv2_150.cvnets_in1k","title":"MobileViTV2 1.5×","source":"timm/mobilevitv2_150.cvnets_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilevitv2_150.cvnets_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilevitv2_150.cvnets_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.27896,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.28204,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":780.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.07,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.86,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.65,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.13,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"2efb0b0d4fabd7b3","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mobilevitv2_175.cvnets_in1k","title":"MobileViTV2 1.75×","source":"timm/mobilevitv2_175.cvnets_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmobilevitv2_175.cvnets_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mobilevitv2_175.cvnets_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.52883,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.53894,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":653.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.62,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.09,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.69,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.17,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"b4417ef7b9fd655c","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mvitv2_base_cls.fb_inw21k","title":"MViTv2 Base (cls)","source":"timm/mvitv2_base_cls.fb_inw21k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmvitv2_base_cls.fb_inw21k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mvitv2_base_cls.fb_inw21k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.7556,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.84467,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":362.7,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.1,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.58,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"3d3f7791239983f6","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mvitv2_huge_cls.fb_inw21k","title":"MViTv2 Huge (cls)","source":"timm/mvitv2_huge_cls.fb_inw21k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmvitv2_huge_cls.fb_inw21k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mvitv2_huge_cls.fb_inw21k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":14.908,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":15.2018,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":67.1,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":7.99,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":7.47,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"6335675c5822c78e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mvitv2_large.fb_in1k","title":"MViTv2 Large","source":"timm/mvitv2_large.fb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmvitv2_large.fb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mvitv2_large.fb_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.78478,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":11.2787,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":7.922,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":11.5736,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":128,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":88.4625,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.23,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.13,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9c6169509920a464","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mvitv2_large.fb_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":17.7313,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":24.5747,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":17.8283,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":24.6387,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":56.2,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":40.5818,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.23,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.13,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ead2642b98d9a493","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mvitv2_large.fb_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":36.2288,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":36.3262,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":27.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.23,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.13,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"665b3aeaa9fd3fd4","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mvitv2_large.fb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.92739,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":7.08167,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":144.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":85.23,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.13,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.97,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.45,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"485765ed3746bf0c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/mvitv2_large.fb_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.97371,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":9.00818,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":111.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.23,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.13,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1ba1d495603a3a6e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mvitv2_large_cls.fb_inw21k","title":"MViTv2 Large (cls)","source":"timm/mvitv2_large_cls.fb_inw21k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmvitv2_large_cls.fb_inw21k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mvitv2_large_cls.fb_inw21k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.31231,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":7.44366,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":136.7,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.96,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.44,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"1f2850ed342e3813","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/mvitv2_tiny.fb_in1k","title":"MViTv2 Tiny","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fmvitv2_tiny.fb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/mvitv2_tiny.fb_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.223,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.226,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":817.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.18,"unit":"%","headline":true},{"description":"Top-5","value":0.68,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16","title":"Nemotron 3 Nano 4B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=nvidia%2FNVIDIA-Nemotron-3-Nano-4B-BF16&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16","hardware":"thor","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":40.3,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":31.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":8212,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":10375,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":43.4,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":33,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":73.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":73.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":90.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":89.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":83.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":84.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":47.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":46.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.62,"unit":"J","higherIsBetter":false,"digits":2,"baseline":2.03,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":0.62,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":0.49,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":65.5,"unit":"W","higherIsBetter":false,"digits":1,"baseline":64.3,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":78.4,"unit":"W","higherIsBetter":false,"digits":1,"baseline":80.2,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":22.8,"unit":"W","higherIsBetter":false,"digits":1,"baseline":22.5,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":7.26,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":9.73,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":6.75,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":5.74,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"94b4147be37a3f61","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16","hardware":"thor","variant":"huggingface_fp16_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":31.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":10375,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":33,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":73.4,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"GSM8K","value":89.1,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"IFEval","value":84.3,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"MMLU-Pro","value":46.8,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":2.03,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.49,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":64.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":80.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":22.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":9.73,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16","hardware":"thor","variant":"huggingface_int4_embedl_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":40.3,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":31.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":8212,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":10375,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":43.4,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":33,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":73.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":73.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":90.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":89.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":83.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":84.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":47.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":46.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.62,"unit":"J","higherIsBetter":false,"digits":2,"baseline":2.03,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":0.62,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":0.49,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":65.5,"unit":"W","higherIsBetter":false,"digits":1,"baseline":64.3,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":78.4,"unit":"W","higherIsBetter":false,"digits":1,"baseline":80.2,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":22.8,"unit":"W","higherIsBetter":false,"digits":1,"baseline":22.5,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":7.26,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":9.73,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":6.75,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":5.74,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"94b4147be37a3f61","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16","hardware":"orin","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":22.1,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":1474,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":28.8,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":73.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":73.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":90.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":89.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":83.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":84.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":47.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":46.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.83,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.55,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":40.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":45.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.1,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":1.67,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":6.66,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":5.74,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"05cdff8c9b88375e","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16","hardware":"orin","variant":"huggingface_int4_embedl_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":22.1,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":1474,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":28.8,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":73.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":73.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":90.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":89.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":83.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":84.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":47.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":46.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.83,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.55,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":40.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":45.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.1,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":1.67,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":6.66,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":5.74,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"05cdff8c9b88375e","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-BF16","title":"Nemotron 3.5 Bolt E4B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=nvidia%2FNVIDIA-Nemotron-3.5-Bolt-E4B-BF16&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-BF16","hardware":"thor","metrics":[{"title":"Throughput","metrics":[{"description":"Decode","value":80.68,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":24.93,"baselineLabel":"Baseline FP16 (NVIDIA BF16 weights)","headline":true},{"description":"End-to-end","value":64.84,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":24.16,"baselineLabel":"Baseline FP16 (NVIDIA BF16 weights)"},{"description":"Prompt processing","value":5124,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":10076,"baselineLabel":"Baseline FP16 (NVIDIA BF16 weights)"}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.0423,"unit":"J","higherIsBetter":false,"digits":2,"baseline":2.8023,"baselineLabel":"Baseline FP16 (NVIDIA BF16 weights)"},{"description":"Generated tokens per joule","value":0.959,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":0.357,"baselineLabel":"Baseline FP16 (NVIDIA BF16 weights)"},{"description":"Power while generating, mean","value":67.586,"unit":"W","higherIsBetter":false,"digits":1,"baseline":67.694,"baselineLabel":"Baseline FP16 (NVIDIA BF16 weights)"},{"description":"Power while generating, peak","value":83.217,"unit":"W","higherIsBetter":false,"digits":1,"baseline":88.174,"baselineLabel":"Baseline FP16 (NVIDIA BF16 weights)"},{"description":"Power at idle","value":22.422,"unit":"W","higherIsBetter":false,"digits":1,"baseline":22.534,"baselineLabel":"Baseline FP16 (NVIDIA BF16 weights)"}]},{"title":"Size","metrics":[{"description":"Model size","value":8.62,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":15.25,"baselineLabel":"Baseline FP16 (NVIDIA BF16 weights)"}]}],"labels":["Text","LLM","Generation"],"build":"2026110307432ffe","updatedAt":"2026-11-04T21:55:47.000Z"},{"model":"nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-BF16","hardware":"thor","variant":"huggingface_fp16_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"Decode","value":24.93,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"End-to-end","value":24.16,"unit":"tok/s","higherIsBetter":true,"digits":1},{"description":"Prompt processing","value":10076,"unit":"tok/s","higherIsBetter":true,"digits":0}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":2.8023,"unit":"J","higherIsBetter":false,"digits":2},{"description":"Generated tokens per joule","value":0.357,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":67.694,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":88.174,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":22.534,"unit":"W","higherIsBetter":false,"digits":1}]},{"title":"Size","metrics":[{"description":"Model size","value":15.25,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"202610291004f60f","updatedAt":"2026-11-04T21:55:47.000Z"},{"model":"nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-BF16","hardware":"orin","metrics":[{"title":"Throughput","metrics":[{"description":"Decode, 3,072 in / 50 out","value":46,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"End-to-end","value":36.76,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":14.88,"baselineLabel":"Baseline FP16 (NVIDIA BF16 weights)"},{"description":"Prompt processing","value":2650,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":3191,"baselineLabel":"Baseline FP16 (NVIDIA BF16 weights)"},{"description":"Decode","value":46.95,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":15.96,"baselineLabel":"Baseline FP16 (NVIDIA BF16 weights)"}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.1,"unit":"J","higherIsBetter":false,"digits":2,"baseline":2.6461,"baselineLabel":"Baseline FP16 (NVIDIA BF16 weights)"},{"description":"Generated tokens per joule","value":0.909,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":0.378,"baselineLabel":"Baseline FP16 (NVIDIA BF16 weights)"},{"description":"Power while generating, mean","value":40.438,"unit":"W","higherIsBetter":false,"digits":1,"baseline":39.383,"baselineLabel":"Baseline FP16 (NVIDIA BF16 weights)"},{"description":"Power while generating, peak","value":55.293,"unit":"W","higherIsBetter":false,"digits":1,"baseline":59.067,"baselineLabel":"Baseline FP16 (NVIDIA BF16 weights)"},{"description":"Power at idle","value":12.243,"unit":"W","higherIsBetter":false,"digits":1,"baseline":12.257,"baselineLabel":"Baseline FP16 (NVIDIA BF16 weights)"}]},{"title":"Size","metrics":[{"description":"Model size","value":8.61,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":15.25,"baselineLabel":"Baseline FP16 (NVIDIA BF16 weights)"}]}],"labels":["Text","LLM","Generation"],"build":"202611030743fd34","updatedAt":"2026-11-04T21:55:47.000Z"},{"model":"nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-BF16","hardware":"orin","variant":"huggingface_fp16_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"Decode","value":15.96,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"End-to-end","value":14.88,"unit":"tok/s","higherIsBetter":true,"digits":1},{"description":"Prompt processing","value":3191,"unit":"tok/s","higherIsBetter":true,"digits":0}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":2.6461,"unit":"J","higherIsBetter":false,"digits":2},{"description":"Generated tokens per joule","value":0.378,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":39.383,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":59.067,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.257,"unit":"W","higherIsBetter":false,"digits":1}]},{"title":"Size","metrics":[{"description":"Model size","value":15.25,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"20261103154073b8","updatedAt":"2026-11-04T21:55:47.000Z"},{"model":"nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-BF16","hardware":"orin-nx-16gb","metrics":[{"title":"Throughput","metrics":[{"description":"Decode, 3,072 in / 50 out","value":29.59,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"End-to-end","value":21.67,"unit":"tok/s","higherIsBetter":true,"digits":1},{"description":"Prompt processing","value":1231,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":29.93,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.1461,"unit":"J","higherIsBetter":false,"digits":2},{"description":"Generated tokens per joule","value":0.873,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":24.835,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":30.544,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":10.408,"unit":"W","higherIsBetter":false,"digits":1}]},{"title":"Size","metrics":[{"description":"Model size","value":8.61,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"2026110307430269","updatedAt":"2026-11-04T16:28:03.000Z"},{"model":"nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-BF16","hardware":"orin-nano-super","metrics":[{"title":"Throughput","metrics":[{"description":"Decode, 3,072 in / 50 out","value":28,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"End-to-end","value":18.91,"unit":"tok/s","higherIsBetter":true,"digits":1},{"description":"Prompt processing","value":890,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":28.41,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.1123,"unit":"J","higherIsBetter":false,"digits":2},{"description":"Generated tokens per joule","value":0.899,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":21.03,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":22.886,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":8.011,"unit":"W","higherIsBetter":false,"digits":1}]},{"title":"Size","metrics":[{"description":"Model size","value":8.61,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"202611030743355c","updatedAt":"2026-11-04T16:28:03.000Z"}]},{"name":"timm/nest_base_jx.goog_in1k","title":"NesT Base (jx)","source":"timm/nest_base_jx.goog_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fnest_base_jx.goog_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/nest_base_jx.goog_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.61157,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.81821,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1.737,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.93372,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":612,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":543.494,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.97,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.09,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"dadb1a651e42ee5b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/nest_base_jx.goog_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.91553,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.92627,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":254,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.97,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.09,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"fd2eda32bb577344","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/nest_base_jx.goog_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.16527,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":8.185,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":121.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.97,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.09,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d45c849df29750d9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/nest_base_jx.goog_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.2681,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.33563,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":440.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.97,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.09,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.28,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.76,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"2566fe0d337dcea1","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/nest_base_jx.goog_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.81211,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.82703,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":355.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.97,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.09,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"05c00fb5322a08bf","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/nest_small_jx.goog_in1k","title":"NesT Small (jx)","source":"timm/nest_small_jx.goog_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fnest_small_jx.goog_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/nest_small_jx.goog_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.39387,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.52814,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":709.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.64,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.99,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1961636e6ad40881","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/nest_small_jx.goog_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.26482,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.27808,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":304.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.64,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.99,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f00aa9dcd64c7cb3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/nest_small_jx.goog_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.62602,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.65491,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":150.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.64,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.99,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"344b62976d5b7c35","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/nest_small_jx.goog_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.72843,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.73352,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":578,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.64,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.99,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.95,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.43,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"4379484c1004ccec","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/nest_small_jx.goog_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.23436,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.24402,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":447.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.64,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.99,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c75960edebad83fb","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/nest_tiny_jx.goog_in1k","title":"NesT Tiny (jx)","source":"timm/nest_tiny_jx.goog_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fnest_tiny_jx.goog_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/nest_tiny_jx.goog_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.695719,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.831333,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.769836,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.910583,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1407,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1168.73,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.03,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.46,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"41b68f26a17e2078","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/nest_tiny_jx.goog_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.92746,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.93616,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":515.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.03,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.46,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c2e5616d9aff0e9c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/nest_tiny_jx.goog_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.00108,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.0235,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":248.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.03,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.46,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2020dbf3a2eac680","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/nest_tiny_jx.goog_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.929898,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.934509,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1073.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.03,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.46,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.72,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.19,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"0d819d9d5bdefd93","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/nest_tiny_jx.goog_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.31304,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.31879,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":761,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.03,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.46,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"33cf6d32501ac152","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/nextvit_base.bd_in1k","title":"Next-ViT Base","source":"timm/nextvit_base.bd_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fnextvit_base.bd_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/nextvit_base.bd_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.37129,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.50665,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":721.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.39,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.57,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ddb62c7214bfab07","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/nextvit_base.bd_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.29529,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.30585,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":301.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.39,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.57,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1d56f95de43add54","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/nextvit_base.bd_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.25006,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.26477,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":159.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.39,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.57,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"74f2a5692d0974ad","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/nextvit_base.bd_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.24725,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.30444,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":444.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.39,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.57,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.15,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.62,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"84551d38293f7764","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/nextvit_base.bd_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.07799,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.0946,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":324.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.39,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.57,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b38b67bd33f57803","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/nextvit_large.bd_in1k","title":"Next-ViT Large","source":"timm/nextvit_large.bd_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fnextvit_large.bd_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/nextvit_large.bd_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.74011,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.87628,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":567.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.66,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"64521669d9ebfa36","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/nextvit_large.bd_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.22373,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.23706,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":235.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.66,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"903c272fcd761233","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/nextvit_large.bd_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.99643,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":8.01257,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":124.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.66,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3f71da97c277a07f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/nextvit_large.bd_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.89892,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.98743,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":344.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.66,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.33,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.81,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e8a6dd240c424e49","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/nextvit_large.bd_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.08537,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.1037,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":244.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.66,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"66d246e15712fa8a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/nextvit_small.bd_in1k","title":"Next-ViT Small","source":"timm/nextvit_small.bd_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fnextvit_small.bd_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/nextvit_small.bd_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.994274,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.09393,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":988.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.55,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.21,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0ffd26404c144f2b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/nextvit_small.bd_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.40528,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.41467,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":413.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.55,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.21,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5cc919881d28a038","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/nextvit_small.bd_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.6028,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.62445,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":216.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.55,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.21,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"885e3b1e66bf2c5a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/nextvit_small.bd_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.54738,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.55078,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":645.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.55,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.21,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.96,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.44,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e01171897ef9d703","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/nextvit_small.bd_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.02209,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.04181,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":494.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.55,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.21,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c88bf485fe392e69","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/nf_regnet_b1.ra2_in1k","title":"NF-RegNet-B1","source":"timm/nf_regnet_b1.ra2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fnf_regnet_b1.ra2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/nf_regnet_b1.ra2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.828581,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.83136,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1204.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.19,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.12,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.66,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.14,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e09ce4b8fdd34833","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/dm_nfnet_f0.dm_in1k","title":"NFNet-F0 (DeepMind)","source":"timm/dm_nfnet_f0.dm_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdm_nfnet_f0.dm_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/dm_nfnet_f0.dm_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.60834,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.68854,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":383.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.37,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.66,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"f3505f94deff3489","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/dm_nfnet_f1.dm_in1k","title":"NFNet-F1 (DeepMind)","source":"timm/dm_nfnet_f1.dm_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdm_nfnet_f1.dm_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/dm_nfnet_f1.dm_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.49628,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.60193,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":181.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.46,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.58,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.36,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.84,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"996aaaad53e2b0a5","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/dm_nfnet_f2.dm_in1k","title":"NFNet-F2 (DeepMind)","source":"timm/dm_nfnet_f2.dm_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdm_nfnet_f2.dm_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/dm_nfnet_f2.dm_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.278,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":8.44354,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":120.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.06,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.85,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.21,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.69,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"3a0d09dd3beb7bd1","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/dm_nfnet_f3.dm_in1k","title":"NFNet-F3 (DeepMind)","source":"timm/dm_nfnet_f3.dm_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdm_nfnet_f3.dm_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/dm_nfnet_f3.dm_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":12.1599,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":12.4609,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":82.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.91,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.27,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4.06,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.54,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"6efcc1517ab9edd9","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/dm_nfnet_f4.dm_in1k","title":"NFNet-F4 (DeepMind)","source":"timm/dm_nfnet_f4.dm_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdm_nfnet_f4.dm_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/dm_nfnet_f4.dm_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":17.5,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":18.0804,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":57.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":85.37,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.42,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4.91,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":4.38,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"23861937786eb16a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/dm_nfnet_f5.dm_in1k","title":"NFNet-F5 (DeepMind)","source":"timm/dm_nfnet_f5.dm_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdm_nfnet_f5.dm_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/dm_nfnet_f5.dm_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":23.1059,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":23.7372,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":43.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":85.62,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.51,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":5.76,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":5.23,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"67ed2c23bb2f978c","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/dm_nfnet_f6.dm_in1k","title":"NFNet-F6 (DeepMind)","source":"timm/dm_nfnet_f6.dm_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdm_nfnet_f6.dm_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/dm_nfnet_f6.dm_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":29.1759,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":29.6796,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":34.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":86,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.72,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":6.6,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":6.08,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"583c15061d108199","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/nfnet_l0.ra2_in1k","title":"NFNet-L0","source":"timm/nfnet_l0.ra2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fnfnet_l0.ra2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/nfnet_l0.ra2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.56551,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.57233,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":638.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.75,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.49,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"29640f254315dba4","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"nvidia/sparse4d","upstream":"nvidia/sparse4d_rn50","upstreamUrl":"https://catalog.ngc.nvidia.com/orgs/nvidia/teams/tao/models/sparse4d_rn50","source":"nvidia/tao/sparse4d_rn50:trainable_v2.2","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=nvidia%2Fsparse4d&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"nvidia/sparse4d","hardware":"thor","precision":"FP8","metrics":[{"title":"Latency","subtitle":"batch 1 = 13 camera views · Baseline TensorRT (FP16) as baseline","metrics":[{"description":"One frame (13 views), median","value":82.0449,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":93.4058,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Throughput","value":12.188448032723546,"unit":"qps","digits":1,"baseline":10.705973290737834,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"PhysicalAI-SmartSpaces Warehouse_015 · 9,000 frames, 13 cameras · HOTA · Baseline TensorRT (FP16) as baseline","metrics":[{"description":"HOTA","value":47.74,"digits":2,"baseline":49.77,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"DetA","value":49.2,"digits":2,"baseline":49.59,"baselineLabel":"Baseline TensorRT"},{"description":"AssA","value":49.4,"digits":2,"baseline":52.75,"baselineLabel":"Baseline TensorRT"}]},{"title":"Size","metrics":[{"description":"Engine size","value":75,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":108.7,"baselineLabel":"Baseline TensorRT","improvementWord":"smaller","headline":true}]}],"labels":["Image","Detection"],"runnable":false,"updatedAt":"2026-11-05T11:06:45.000Z"}]},{"name":"google/owlv2-base-patch16-ensemble","upstream":"google/owlv2-base-patch16-ensemble","title":"OWLv2 Base","upstreamUrl":"https://huggingface.co/google/owlv2-base-patch16-ensemble","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=google%2Fowlv2-base-patch16-ensemble&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"google/owlv2-base-patch16-ensemble","hardware":"thor","precision":"FP16","metrics":[{"title":"Latency","subtitle":"batch 1 · one photo and prompt end to end","metrics":[{"description":"Photo and prompt to boxes, median","value":159.78693349995865,"unit":"ms","higherIsBetter":false,"digits":1,"headline":true},{"description":"Image engine (GPU)","value":13.218624114990234,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Text engine, one query (GPU)","value":0.5801920294761658,"unit":"ms","higherIsBetter":false,"digits":2}]},{"title":"Accuracy","subtitle":"COCO val2017 · 5,000 images · box AP@[.50:.95] · PyTorch FP32 as baseline","metrics":[{"description":"Box AP","value":45.28113766388244,"unit":"%","higherIsBetter":true,"digits":2,"baseline":45.3164484715323,"baselineLabel":"PyTorch FP32","headline":true}]},{"title":"Memory","metrics":[{"description":"Server peak process RSS","value":649.6171875,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]},{"title":"Size","metrics":[{"description":"Engines (image and text)","value":297.8670349121094,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Image","Text","Detection"],"build":"202610172250d094","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"google/owlv2-base-patch16-ensemble","hardware":"thor","variant":"huggingface_fp16_tensorrt","precision":"FP16","metrics":[{"title":"Latency","subtitle":"batch 1 · one photo and prompt end to end","metrics":[{"description":"Photo and prompt to boxes, median","value":159.78693349995865,"unit":"ms","higherIsBetter":false,"digits":1,"headline":true},{"description":"Image engine (GPU)","value":13.218624114990234,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Text engine, one query (GPU)","value":0.5801920294761658,"unit":"ms","higherIsBetter":false,"digits":2}]},{"title":"Accuracy","subtitle":"COCO val2017 · 5,000 images · box AP@[.50:.95] · PyTorch FP32 as baseline","metrics":[{"description":"Box AP","value":45.28113766388244,"unit":"%","higherIsBetter":true,"digits":2,"baseline":45.3164484715323,"baselineLabel":"PyTorch FP32","headline":true}]},{"title":"Memory","metrics":[{"description":"Server peak process RSS","value":649.6171875,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]},{"title":"Size","metrics":[{"description":"Engines (image and text)","value":297.8670349121094,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Image","Text","Detection"],"build":"202610172250d094","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/pit_b_distilled_224.in1k","title":"PiT-B (224, distilled)","source":"timm/pit_b_distilled_224.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fpit_b_distilled_224.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/pit_b_distilled_224.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.04671,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.30209,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1.12164,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.40417,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":935.1,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":754.56,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.93,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.8,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0a79d0cb2b93303b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_b_distilled_224.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.08322,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":7.1084,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":140.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.93,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.8,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c5b72f918c014ab0","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_b_distilled_224.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":11.1218,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":11.1467,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":89.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.93,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.8,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f374545c8e1280d7","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_b_distilled_224.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.25279,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.25775,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":307.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.93,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.8,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.36,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.84,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"1fa6c715bc6681c2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_b_distilled_224.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.46924,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.50092,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":182.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.93,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.8,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"dfdda6c0025a95c1","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/pit_b_224.in1k","title":"PiT-B (224)","source":"timm/pit_b_224.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fpit_b_224.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/pit_b_224.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.03564,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.29219,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1.10724,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.3913,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":944.6,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":756.013,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.46,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.84,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"928ad6c31c90bb08","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_b_224.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.04635,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":7.06995,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":141.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.46,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.84,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a5d956774597baec","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_b_224.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":11.0656,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":11.0902,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":90.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.46,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.84,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7c229c79be3116a5","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_b_224.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.25917,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.26343,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":306.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.46,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.84,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.35,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.83,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"7480fac65c78d035","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_b_224.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.43845,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.4787,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":183.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.46,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.84,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7f9775b81150e113","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/pit_s_distilled_224.in1k","title":"PiT-S (224, distilled)","source":"timm/pit_s_distilled_224.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fpit_s_distilled_224.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/pit_s_distilled_224.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.545357,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.686493,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.619812,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.747742,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1775.7,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1410.35,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.86,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.76,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a55d9813d49a21ae","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_s_distilled_224.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.81887,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.83099,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":260.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.86,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.76,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"37e5c6ffa27693dd","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_s_distilled_224.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.69213,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.71228,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":175,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.86,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.76,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"12a8a6f2bc00f3ed","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_s_distilled_224.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.75357,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.75861,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":569.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.86,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.76,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.8,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.28,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"6ac9c1cc925d7363","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_s_distilled_224.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.09824,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.11542,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":322.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.86,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.76,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2fa0be590c6d1827","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/pit_s_224.in1k","title":"PiT-S (224)","source":"timm/pit_s_224.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fpit_s_224.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/pit_s_224.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.546847,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.673332,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.617554,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.737976,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1780,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1431.84,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.15,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d6945ff52a124037","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_s_224.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.82167,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.83374,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":260.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.15,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"909e6c17dfd0f8d2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_s_224.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.6741,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.69147,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":175.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.15,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"042ab04eca6f4f25","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_s_224.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.72678,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.73199,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":578.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.15,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.79,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.27,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"114d8f1da81b8c43","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_s_224.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.07299,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.0863,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":325.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.15,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ee1a6de6939c35f3","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/pit_ti_distilled_224.in1k","title":"PiT-Ti (224, distilled)","source":"timm/pit_ti_distilled_224.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fpit_ti_distilled_224.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/pit_ti_distilled_224.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.405005,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.492249,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2439.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":73.87,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.59,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ed54b8b55eb9bc34","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_ti_distilled_224.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.01098,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.02271,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":494.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":73.87,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.59,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0d4439a3637c117d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_ti_distilled_224.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.80478,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.81256,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":354.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":73.87,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.59,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"cdb23161e2ac603d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_ti_distilled_224.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.825224,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.827393,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1209.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":73.87,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.59,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.58,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":60.2,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"ca6565195f8feab4","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_ti_distilled_224.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.48995,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.49506,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":670.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":73.87,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.59,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6adec57072f418e9","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/pit_ti_224.in1k","title":"PiT-Ti (224)","source":"timm/pit_ti_224.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fpit_ti_224.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/pit_ti_224.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.402669,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.475281,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2432.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.06,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a16b25e6b99c4588","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_ti_224.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.97483,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.98132,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":503.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.06,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"addf2d264866c41e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_ti_224.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.78705,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.79626,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":356.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.06,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2bb66f8d765cb046","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_ti_224.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.81558,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.817322,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1224.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":72.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.06,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.58,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":57.2,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"65f43f0f15bad0fe","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_ti_224.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.47138,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.48083,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":679.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.06,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ba404ca893cac09e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/pit_xs_distilled_224.in1k","title":"PiT-XS (224, distilled)","source":"timm/pit_xs_distilled_224.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fpit_xs_distilled_224.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/pit_xs_distilled_224.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.474788,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.553345,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2054.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.25,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1265269e09fcae7c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_xs_distilled_224.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.70451,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.71356,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":367.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.25,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9fd859a5d49b4018","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_xs_distilled_224.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.89758,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.9093,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":255.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.25,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"40d1ef2c9cda9b8f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_xs_distilled_224.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.20676,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.20844,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":827.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.25,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.65,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.13,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"c48aa28563269584","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_xs_distilled_224.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.15218,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.15509,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":464.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.25,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d55290cedc005ec9","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/pit_xs_224.in1k","title":"PiT-XS (224)","source":"timm/pit_xs_224.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fpit_xs_224.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/pit_xs_224.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.471966,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.541992,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2061.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.02,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.02,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"26e3c371b35571bd","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_xs_224.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.70793,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.71777,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":367.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.02,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.02,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f4e5bdd0d4d0437b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_xs_224.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.88068,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.89392,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":256.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.02,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.02,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1cfc5291a623dd01","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_xs_224.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.19713,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.19904,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":834.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.02,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.02,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.64,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.12,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e3cb4fca6dc1448b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/pit_xs_224.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.12962,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.13806,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":469.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.02,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.02,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"cf4ec2ce3711fe09","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/poolformer_m36.sail_in1k","title":"PoolFormer-M36","source":"timm/poolformer_m36.sail_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fpoolformer_m36.sail_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/poolformer_m36.sail_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.60386,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.73181,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":274.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.46,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.5,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9525bcdfe7168b7c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/poolformer_m36.sail_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.3159,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.33179,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":187.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.46,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.5,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0612edbd906fd2f7","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/poolformer_m36.sail_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":10.6461,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":10.6741,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":93.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.46,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.5,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8322d2cfd2532841","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/poolformer_m36.sail_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.09204,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.14661,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":477.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.46,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.5,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.17,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.64,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"119b1367b56fbab4","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/poolformer_m36.sail_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.9872,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.99213,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":334.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.46,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.5,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8e10e15a6412aa45","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/poolformer_m48.sail_in1k","title":"PoolFormer-M48","source":"timm/poolformer_m48.sail_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fpoolformer_m48.sail_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/poolformer_m48.sail_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.74935,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.81726,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":209.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.99,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.69,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b6bd5e10f36b1b8e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/poolformer_m48.sail_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.95494,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.97937,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":143,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.99,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.69,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"293f18aa06be17e8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/poolformer_m48.sail_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":13.8865,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":13.9094,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":71.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.99,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.69,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a723d0017076377f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/poolformer_m48.sail_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.66282,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.7193,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":375.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.99,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.69,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.36,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.84,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"95fb5e272aa624a0","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/poolformer_m48.sail_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.83689,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.8526,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":260.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.99,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.69,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2c9e54247d5f22e5","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/poolformer_s12.sail_in1k","title":"PoolFormer-S12","source":"timm/poolformer_s12.sail_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fpoolformer_s12.sail_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/poolformer_s12.sail_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.641699,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.643005,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1554.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":75.06,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.33,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.66,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.14,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"a362489c4e4cbb33","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/poolformer_s24.sail_in1k","title":"PoolFormer-S24","source":"timm/poolformer_s24.sail_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fpoolformer_s24.sail_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/poolformer_s24.sail_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.8174,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.06044,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1.9223,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.16534,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":543.5,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":479.742,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.45,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.67,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8f8f2f38f90fbc24","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/poolformer_s24.sail_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.01331,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.02643,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":330.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.45,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.67,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1f4770ee56210265","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/poolformer_s24.sail_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.6444,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.65607,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":176.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.45,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.67,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"032e7e2c0136b097","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/poolformer_s24.sail_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.09194,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.09454,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":914.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.45,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.67,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.77,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.25,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"da1e2c732d9cf6a9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/poolformer_s24.sail_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.82141,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.82904,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":548.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.45,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.67,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b262bbda540596a4","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/poolformer_s36.sail_in1k","title":"PoolFormer-S36","source":"timm/poolformer_s36.sail_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fpoolformer_s36.sail_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/poolformer_s36.sail_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.81937,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.90857,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":351.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.73,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.14,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9cf565ede58bf672","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/poolformer_s36.sail_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.35116,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.36621,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":228.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.73,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.14,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"72d99d8d91bad8f5","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/poolformer_s36.sail_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.12664,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":8.14343,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":122.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.73,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.14,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9adf483fc2359dfd","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/poolformer_s36.sail_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.65033,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.65311,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":605.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.73,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.14,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.88,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.36,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"b7af01520e09ab82","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/poolformer_s36.sail_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.52608,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.52954,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":395.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.73,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.14,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a3a0409643c84b1e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/poolformerv2_s12.sail_in1k","title":"PoolFormerV2-S12","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fpoolformerv2_s12.sail_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/poolformerv2_s12.sail_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":547.913,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":549.011,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":1820.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":60.15,"unit":"%","headline":true},{"description":"Top-5","value":82.38,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/pvt_v2_b0.in1k","title":"PVT v2-B0","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fpvt_v2_b0.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/pvt_v2_b0.in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":803.345,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":806.152,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":1242.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":48.45,"unit":"%","headline":true},{"description":"Top-5","value":74.4,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/pvt_v2_b3.in1k","title":"PVT v2-B3","source":"timm/pvt_v2_b3.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fpvt_v2_b3.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/pvt_v2_b3.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.15403,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.19617,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":240.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.71,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.71,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.06,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.54,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"db714abb71487752","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/pvt_v2_b4.in1k","title":"PVT v2-B4","source":"timm/pvt_v2_b4.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fpvt_v2_b4.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/pvt_v2_b4.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.99803,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.1225,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":166.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.22,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.11,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.27,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.75,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"0c0b54d8f452a8ca","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/pvt_v2_b5.in1k","title":"PVT v2-B5","source":"timm/pvt_v2_b5.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fpvt_v2_b5.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/pvt_v2_b5.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.58113,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.65942,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":151.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.36,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.09,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.5,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.98,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"678df811f6d968d2","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"Qwen/Qwen3-0.6B","title":"Qwen3 0.6B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=Qwen%2FQwen3-0.6B&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"Qwen/Qwen3-0.6B","hardware":"thor","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":266,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":151.6,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":34420,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":68495,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":301.3,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":156,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":46.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":49.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":61.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":64.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":53.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":58.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":25.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":26.9,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.2,"unit":"J","higherIsBetter":false,"digits":2,"baseline":0.37,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":4.96,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":2.67,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":53.6,"unit":"W","higherIsBetter":false,"digits":1,"baseline":56.8,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":58.2,"unit":"W","higherIsBetter":false,"digits":1,"baseline":60.4,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":21.6,"unit":"W","higherIsBetter":false,"digits":1,"baseline":24.8,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":2.42,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":3.14,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":1.86,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":0.85,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"84cdc3713a399393","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-0.6B","hardware":"thor","variant":"huggingface_fp16_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":151.6,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":68495,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":156,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":49.8,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"GSM8K","value":64.4,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"IFEval","value":58.2,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"MMLU-Pro","value":26.9,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.37,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":2.67,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":56.8,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":60.4,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":24.8,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":3.14,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-0.6B","hardware":"thor","variant":"huggingface_int4_embedl-flashhead_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":266,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":151.6,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":34420,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":68495,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":301.3,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":156,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":46.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":49.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":61.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":64.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":53.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":58.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":25.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":26.9,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.2,"unit":"J","higherIsBetter":false,"digits":2,"baseline":0.37,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":4.96,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":2.67,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":53.6,"unit":"W","higherIsBetter":false,"digits":1,"baseline":56.8,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":58.2,"unit":"W","higherIsBetter":false,"digits":1,"baseline":60.4,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":21.6,"unit":"W","higherIsBetter":false,"digits":1,"baseline":24.8,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":2.42,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":3.14,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":1.86,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":0.85,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"84cdc3713a399393","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-0.6B","hardware":"orin","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":155.3,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":95.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":18618,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":21400,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":178,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":101.7,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":46.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":49.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":60.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":64.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":53.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":58.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":24.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":26.9,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.21,"unit":"J","higherIsBetter":false,"digits":2,"baseline":0.39,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":4.78,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":2.54,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":32.5,"unit":"W","higherIsBetter":false,"digits":1,"baseline":37.4,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":36.7,"unit":"W","higherIsBetter":false,"digits":1,"baseline":41.6,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":11.5,"unit":"W","higherIsBetter":false,"digits":1,"baseline":11.9,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":0.81,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":1.68,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":1.76,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":0.84,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"47f424a7c5d33e15","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-0.6B","hardware":"orin","variant":"huggingface_fp16_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":95.1,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":21400,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":101.7,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":49.8,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"GSM8K","value":64.4,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"IFEval","value":58.2,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"MMLU-Pro","value":26.9,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.39,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":2.54,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":37.4,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":41.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":11.9,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":1.68,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-0.6B","hardware":"orin","variant":"huggingface_int4_embedl-flashhead_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":155.3,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":95.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":18618,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":21400,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":178,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":101.7,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":46.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":49.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":60.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":64.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":53.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":58.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":24.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":26.9,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.21,"unit":"J","higherIsBetter":false,"digits":2,"baseline":0.39,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":4.78,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":2.54,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":32.5,"unit":"W","higherIsBetter":false,"digits":1,"baseline":37.4,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":36.7,"unit":"W","higherIsBetter":false,"digits":1,"baseline":41.6,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":11.5,"unit":"W","higherIsBetter":false,"digits":1,"baseline":11.9,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":0.81,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":1.68,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":1.76,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":0.84,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"47f424a7c5d33e15","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-0.6B","hardware":"orin-nx-16gb","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":70.4,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":7201,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":82.8,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":47,"unit":"%","higherIsBetter":true,"digits":1,"baseline":49.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":62.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":64.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":54.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":58.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":24,"unit":"%","higherIsBetter":true,"digits":1,"baseline":26.9,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.26,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":3.82,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":18.4,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":20,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":8.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":0.84,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"75195b85f194a022","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-0.6B","hardware":"orin-nx-16gb","variant":"huggingface_int4_embedl_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":70.4,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":7201,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":82.8,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":47,"unit":"%","higherIsBetter":true,"digits":1,"baseline":49.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":62.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":64.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":54.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":58.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":24,"unit":"%","higherIsBetter":true,"digits":1,"baseline":26.9,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.26,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":3.82,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":18.4,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":20,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":8.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":0.84,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"75195b85f194a022","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-0.6B","hardware":"orin-nano-super","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":88.8,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":6759,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":111.6,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":47.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":49.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":63.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":64.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":54.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":58.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":24.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":26.9,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.21,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":4.8,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":18.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":20,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":7.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":1.43,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":1.76,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":0.84,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"3727111af19fef8d","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-0.6B","hardware":"orin-nano-super","variant":"huggingface_int4_embedl-flashhead_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":88.8,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":6759,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":111.6,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":47.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":49.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":63.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":64.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":54.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":58.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":24.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":26.9,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.21,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":4.8,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":18.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":20,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":7.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":1.43,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":1.76,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":0.84,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"3727111af19fef8d","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"Qwen/Qwen3-1.7B","title":"Qwen3 1.7B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=Qwen%2FQwen3-1.7B&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"Qwen/Qwen3-1.7B","hardware":"thor","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":117.8,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":63.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":13717,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":31508,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":135.5,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":65.4,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":63.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":81.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":83,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":66.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":68.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":42.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":46,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.53,"unit":"J","higherIsBetter":false,"digits":2,"baseline":0.94,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":1.87,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":1.07,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":63,"unit":"W","higherIsBetter":false,"digits":1,"baseline":59.7,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":70.3,"unit":"W","higherIsBetter":false,"digits":1,"baseline":68.9,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":22.7,"unit":"W","higherIsBetter":false,"digits":1,"baseline":23.6,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":3.43,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":5.63,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":2.89,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":1.88,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"60d07f7b71b606f3","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-1.7B","hardware":"thor","variant":"huggingface_fp16_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":63.8,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":31508,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":65.4,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":65.8,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"GSM8K","value":83,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"IFEval","value":68.2,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"MMLU-Pro","value":46,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.94,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.07,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":59.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":68.9,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":23.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":5.63,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-1.7B","hardware":"thor","variant":"huggingface_int4_embedl_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":117.8,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":63.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":13717,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":31508,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":135.5,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":65.4,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":63.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":81.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":83,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":66.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":68.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":42.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":46,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.53,"unit":"J","higherIsBetter":false,"digits":2,"baseline":0.94,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":1.87,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":1.07,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":63,"unit":"W","higherIsBetter":false,"digits":1,"baseline":59.7,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":70.3,"unit":"W","higherIsBetter":false,"digits":1,"baseline":68.9,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":22.7,"unit":"W","higherIsBetter":false,"digits":1,"baseline":23.6,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":3.43,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":5.63,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":2.89,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":1.88,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"60d07f7b71b606f3","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-1.7B","hardware":"orin","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":74.8,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":8346,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":86.6,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":63.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":81.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":83,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":66.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":68.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":42.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":46,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.54,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.87,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":40,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":47.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.1,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":0.93,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":2.79,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":1.87,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"65e397178fa852e6","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-1.7B","hardware":"orin","variant":"huggingface_int4_embedl_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":74.8,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":8346,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":86.6,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":63.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":81.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":83,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":66.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":68.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":42.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":46,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.54,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.87,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":40,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":47.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.1,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":0.93,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":2.79,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":1.87,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"65e397178fa852e6","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-1.7B","hardware":"orin-nx-16gb","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":37.4,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":3218,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":45.5,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":63.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":82,"unit":"%","higherIsBetter":true,"digits":1,"baseline":83,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":64.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":68.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":42.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":46,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.54,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.86,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":20.1,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":22.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":8.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":2.79,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":1.87,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-1.7B","hardware":"orin-nx-16gb","variant":"huggingface_int4_embedl_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":37.4,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":3218,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":45.5,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":63.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":82,"unit":"%","higherIsBetter":true,"digits":1,"baseline":83,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":64.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":68.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":42.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":46,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.54,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.86,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":20.1,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":22.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":8.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":2.79,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":1.87,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-1.7B","hardware":"orin-nano-super","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":38.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":2832,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":49.1,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":62.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":80.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":83,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":65.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":68.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":43.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":46,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.55,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.82,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":21.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":22.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":7.9,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":2.6,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":2.79,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":1.87,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"62b2da11a3b3aaf7","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-1.7B","hardware":"orin-nano-super","variant":"huggingface_int4_embedl_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":38.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":2832,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":49.1,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":62.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":80.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":83,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":65.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":68.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":43.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":46,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.55,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.82,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":21.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":22.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":7.9,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":2.6,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":2.79,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":1.87,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"62b2da11a3b3aaf7","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"Qwen/Qwen3-30B-A3B","title":"Qwen3 30B-A3B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=Qwen%2FQwen3-30B-A3B&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"Qwen/Qwen3-30B-A3B","hardware":"thor","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":67.6,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":4110,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":91.1,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":82.8,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"GSM8K","value":94.2,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"IFEval","value":81,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"MMLU-Pro","value":73.2,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.92,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.09,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":61.9,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":74.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":21.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":17.46,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":16.66,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":15.65,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"5085acce40b866d8","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-30B-A3B","hardware":"thor","variant":"huggingface_int4_embedl-flashhead_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":67.6,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":4110,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":91.1,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":82.8,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"GSM8K","value":94.2,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"IFEval","value":81,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"MMLU-Pro","value":73.2,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.92,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.09,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":61.9,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":74.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":21.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":17.46,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":16.66,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":15.65,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"5085acce40b866d8","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"Qwen/Qwen3-4B","title":"Qwen3 4B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=Qwen%2FQwen3-4B&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"Qwen/Qwen3-4B","hardware":"thor","variant":"huggingface_int4_modelopt_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":62.58,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":5596,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":75.63,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"MMLU-mini","value":64.5,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.156,"unit":"J","higherIsBetter":false,"digits":3,"headline":true},{"description":"Generated tokens per joule","value":0.865,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":72.345,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":85.851,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":22.316,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":5.131,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":4.25,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":3.24,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"1c7046f5af7cfd61","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-4B","hardware":"orin","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":39.28,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":3446,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":47.68,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"MMLU-mini","value":64.5,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.1372,"unit":"J","higherIsBetter":false,"digits":3,"headline":true},{"description":"Generated tokens per joule","value":0.879,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":44.674,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":58.754,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.768,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":1.661,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":4.15,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":3.23,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"369ddd2f6657255e","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-4B","hardware":"orin","variant":"huggingface_int4_modelopt_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":39.28,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":3446,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":47.68,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"MMLU-mini","value":64.5,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.1372,"unit":"J","higherIsBetter":false,"digits":3,"headline":true},{"description":"Generated tokens per joule","value":0.879,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":44.674,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":58.754,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.768,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":1.661,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":4.15,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":3.23,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"369ddd2f6657255e","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-4B","hardware":"orin-nx-16gb","variant":"huggingface_int4_modelopt_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":19.67,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":1300,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":25.75,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"MMLU-mini","value":64.5,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.0413,"unit":"J","higherIsBetter":false,"digits":3,"headline":true},{"description":"Generated tokens per joule","value":0.96,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":20.479,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":23.249,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":8.656,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":4.607,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":4.15,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":3.23,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"Qwen/Qwen3-8B","title":"Qwen3 8B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=Qwen%2FQwen3-8B&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"Qwen/Qwen3-8B","hardware":"thor","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":56.4,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":15.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":3242,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":6924,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":77.5,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":15.5,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":80.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":81.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":93.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":93.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":83,"unit":"%","higherIsBetter":true,"digits":1,"baseline":81.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":65,"unit":"%","higherIsBetter":true,"digits":1,"baseline":68.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.49,"unit":"J","higherIsBetter":false,"digits":2,"baseline":4.11,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":0.67,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":0.24,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":84.1,"unit":"W","higherIsBetter":false,"digits":1,"baseline":62,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":95.4,"unit":"W","higherIsBetter":false,"digits":1,"baseline":99.3,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":23.4,"unit":"W","higherIsBetter":false,"digits":1,"baseline":22.5,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":10.27,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":17.41,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":8.68,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":7.67,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"9d3a10b092e65234","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-8B","hardware":"thor","variant":"huggingface_fp16_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":15.1,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":6924,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":15.5,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":81.3,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"GSM8K","value":93.8,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"IFEval","value":81.7,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"MMLU-Pro","value":68.4,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":4.11,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.24,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":62,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":99.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":22.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":17.41,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-8B","hardware":"thor","variant":"huggingface_int4_embedl-dflash_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":56.4,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":15.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":3242,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":6924,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":77.5,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":15.5,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":80.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":81.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":93.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":93.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":83,"unit":"%","higherIsBetter":true,"digits":1,"baseline":81.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":65,"unit":"%","higherIsBetter":true,"digits":1,"baseline":68.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.49,"unit":"J","higherIsBetter":false,"digits":2,"baseline":4.11,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":0.67,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":0.24,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":84.1,"unit":"W","higherIsBetter":false,"digits":1,"baseline":62,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":95.4,"unit":"W","higherIsBetter":false,"digits":1,"baseline":99.3,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":23.4,"unit":"W","higherIsBetter":false,"digits":1,"baseline":22.5,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":10.27,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":17.41,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":8.68,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":7.67,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"9d3a10b092e65234","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-8B","hardware":"orin","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":38.1,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":1997,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":54.4,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":80.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":81.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":93.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":93.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":83,"unit":"%","higherIsBetter":true,"digits":1,"baseline":81.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":65,"unit":"%","higherIsBetter":true,"digits":1,"baseline":68.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.38,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.73,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":52.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":62.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":13,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":7.78,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":8.56,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":7.64,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"635afb56c2a5fa45","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-8B","hardware":"orin","variant":"huggingface_int4_embedl-dflash_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":38.1,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":1997,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":54.4,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":80.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":81.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":93.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":93.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":83,"unit":"%","higherIsBetter":true,"digits":1,"baseline":81.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":65,"unit":"%","higherIsBetter":true,"digits":1,"baseline":68.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.38,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.73,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":52.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":62.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":13,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":7.78,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":8.56,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":7.64,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Text","LLM","Generation"],"build":"635afb56c2a5fa45","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/qwen3_vit_306m.qwen3_5_2b","title":"Qwen3 ViT 306M","source":"timm/qwen3_vit_306m.qwen3_5_2b","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fqwen3_vit_306m.qwen3_5_2b&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/qwen3_vit_306m.qwen3_5_2b","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":173.734,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":175.23,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":5.8,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.96,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.44,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"e4b4899dc2dc2009","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/qwen3_vit_416m.qwen3_5_9b","title":"Qwen3 ViT 416M","source":"timm/qwen3_vit_416m.qwen3_5_9b","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fqwen3_vit_416m.qwen3_5_9b&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/qwen3_vit_416m.qwen3_5_9b","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":228.59,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":236.621,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":4.4,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.84,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.32,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"d24ad03b4d663462","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/qwen3_vit_88m.qwen3_5_0_8b","title":"Qwen3 ViT 88M","source":"timm/qwen3_vit_88m.qwen3_5_0_8b","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fqwen3_vit_88m.qwen3_5_0_8b&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/qwen3_vit_88m.qwen3_5_0_8b","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":62.9371,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":63.8925,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":15.9,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.22,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.7,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"c1680c616eac2db4","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"Qwen/Qwen3-VL-2B-Instruct","title":"Qwen3-VL 2B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=Qwen%2FQwen3-VL-2B-Instruct&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"Qwen/Qwen3-VL-2B-Instruct","hardware":"thor","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":117.5,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":63.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":13727,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":32151,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":135.2,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":65.3,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":63.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":82.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":83,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":64.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":68.9,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":43.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":45.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.53,"unit":"J","higherIsBetter":false,"digits":2,"baseline":0.94,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":1.87,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":1.07,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":62.7,"unit":"W","higherIsBetter":false,"digits":1,"baseline":59.8,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":70.2,"unit":"W","higherIsBetter":false,"digits":1,"baseline":68.5,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":22.4,"unit":"W","higherIsBetter":false,"digits":1,"baseline":22,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":4.43,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":6.41,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":3.65,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":2.64,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"ca575d98750befdc","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-VL-2B-Instruct","hardware":"thor","variant":"huggingface_fp16_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":63.8,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":32151,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":65.3,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":65.8,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"GSM8K","value":83,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"IFEval","value":68.9,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"MMLU-Pro","value":45.5,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.94,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.07,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":59.8,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":68.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":22,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":6.41,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-VL-2B-Instruct","hardware":"thor","variant":"huggingface_int4_embedl_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":117.5,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":63.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":13727,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":32151,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":135.2,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":65.3,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":63.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":82.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":83,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":64.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":68.9,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":43.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":45.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.53,"unit":"J","higherIsBetter":false,"digits":2,"baseline":0.94,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":1.87,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":1.07,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":62.7,"unit":"W","higherIsBetter":false,"digits":1,"baseline":59.8,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":70.2,"unit":"W","higherIsBetter":false,"digits":1,"baseline":68.5,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":22.4,"unit":"W","higherIsBetter":false,"digits":1,"baseline":22,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":4.43,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":6.41,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":3.65,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":2.64,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"ca575d98750befdc","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-VL-2B-Instruct","hardware":"orin","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":74.9,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":8285,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":87,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":63.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":81.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":83,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":66.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":68.9,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":43.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":45.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.54,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.84,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":40.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":48.9,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":1,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":3.56,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":2.64,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"18cce3e1a66e3e03","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-VL-2B-Instruct","hardware":"orin","variant":"huggingface_int4_embedl_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":74.9,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":8285,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":87,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":63.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":81.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":83,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":66.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":68.9,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":43.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":45.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.54,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.84,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":40.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":48.9,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":1,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":3.56,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":2.64,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"18cce3e1a66e3e03","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-VL-2B-Instruct","hardware":"orin-nx-16gb","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":36.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":3179,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":44.7,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":63.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":82.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":83,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":65.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":68.9,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":42.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":45.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.57,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.77,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":20.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":24.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":9.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":2.82,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"c70d9217f0c901dc","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-VL-2B-Instruct","hardware":"orin-nx-16gb","variant":"huggingface_int4_embedl_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":36.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":3179,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":44.7,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":63.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":82.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":83,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":65.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":68.9,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":42.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":45.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.57,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.77,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":20.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":24.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":9.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":2.82,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"c70d9217f0c901dc","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-VL-2B-Instruct","hardware":"orin-nano-super","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":38.6,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":2822,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":49.1,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":65,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":82.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":83,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":68,"unit":"%","higherIsBetter":true,"digits":1,"baseline":68.9,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":44.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":45.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.55,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.81,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":21.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":22.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":8,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":3.65,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":3.56,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":2.64,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"66e7353f68849cea","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-VL-2B-Instruct","hardware":"orin-nano-super","variant":"huggingface_int4_embedl_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":38.6,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":2822,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":49.1,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":65,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":82.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":83,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":68,"unit":"%","higherIsBetter":true,"digits":1,"baseline":68.9,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":44.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":45.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.55,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.81,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":21.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":22.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":8,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":3.65,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":3.56,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":2.64,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"66e7353f68849cea","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"Qwen/Qwen3-VL-4B-Instruct","title":"Qwen3-VL 4B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=Qwen%2FQwen3-VL-4B-Instruct&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"Qwen/Qwen3-VL-4B-Instruct","hardware":"thor","variant":"huggingface_int4_modelopt_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":62.41,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":5586,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":75.41,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"MMLU-mini","value":73.5,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.14,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.877,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":71.148,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":84.614,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":22.309,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":6.146,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":5.03,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":4.02,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"0d3371b1c6824482","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-VL-4B-Instruct","hardware":"orin","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":39.39,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":3469,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":47.76,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"MMLU-mini","value":73.5,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.1201,"unit":"J","higherIsBetter":false,"digits":3,"headline":true},{"description":"Generated tokens per joule","value":0.893,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":44.121,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":58.275,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.602,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":2.605,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":4.94,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":4.02,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"851dcac34cbf0b13","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-VL-4B-Instruct","hardware":"orin","variant":"huggingface_int4_modelopt_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":39.39,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":3469,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":47.76,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"MMLU-mini","value":73.5,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.1201,"unit":"J","higherIsBetter":false,"digits":3,"headline":true},{"description":"Generated tokens per joule","value":0.893,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":44.121,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":58.275,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.602,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":2.605,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":4.94,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":4.02,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"851dcac34cbf0b13","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-VL-4B-Instruct","hardware":"orin-nx-16gb","variant":"huggingface_int4_modelopt_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":19.79,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":1310,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":25.89,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"MMLU-mini","value":73.5,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.058,"unit":"J","higherIsBetter":false,"digits":3,"headline":true},{"description":"Generated tokens per joule","value":0.945,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":20.937,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":23.743,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":8.665,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":5.388,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":4.94,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":4.02,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"Qwen/Qwen3-VL-8B-Instruct","title":"Qwen3-VL 8B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=Qwen%2FQwen3-VL-8B-Instruct&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"Qwen/Qwen3-VL-8B-Instruct","hardware":"thor","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":37.8,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":15.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":3245,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":6485,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":46,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":15.7,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":81.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":82.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":94.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":94.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":83.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":83.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":66.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":70,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.89,"unit":"J","higherIsBetter":false,"digits":2,"baseline":4.16,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":0.53,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":0.24,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":71.3,"unit":"W","higherIsBetter":false,"digits":1,"baseline":63.5,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":91.9,"unit":"W","higherIsBetter":false,"digits":1,"baseline":96.6,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":22.6,"unit":"W","higherIsBetter":false,"digits":1,"baseline":22.1,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":9.05,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":18.97,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":7.79,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":6.78,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"df2f4d2d4e3f6774","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-VL-8B-Instruct","hardware":"thor","variant":"huggingface_fp16_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":15.3,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":6485,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":15.7,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":82.7,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"GSM8K","value":94.5,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"IFEval","value":83.7,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"MMLU-Pro","value":70,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":4.16,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.24,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":63.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":96.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":22.1,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":18.97,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-VL-8B-Instruct","hardware":"thor","variant":"huggingface_int4_embedl_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":37.8,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":15.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":3245,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":6485,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":46,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":15.7,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":81.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":82.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":94.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":94.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":83.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":83.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":66.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":70,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.89,"unit":"J","higherIsBetter":false,"digits":2,"baseline":4.16,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":0.53,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":0.24,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":71.3,"unit":"W","higherIsBetter":false,"digits":1,"baseline":63.5,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":91.9,"unit":"W","higherIsBetter":false,"digits":1,"baseline":96.6,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":22.6,"unit":"W","higherIsBetter":false,"digits":1,"baseline":22.1,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":9.05,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":18.97,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":7.79,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":6.78,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"df2f4d2d4e3f6774","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-VL-8B-Instruct","hardware":"orin","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":25.1,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":2077,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":30.9,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":81.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":82.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":94.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":94.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":83.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":83.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":66.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":70,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.87,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.54,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":46.9,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":63,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":5.89,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":7.7,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":6.78,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"988c0e97332a2129","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3-VL-8B-Instruct","hardware":"orin","variant":"huggingface_int4_embedl_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":25.1,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":2077,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":30.9,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":81.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":82.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":94.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":94.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":83.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":83.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":66.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":70,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.87,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.54,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":46.9,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":63,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":5.89,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":7.7,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":6.78,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"988c0e97332a2129","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"Qwen/Qwen3.5-0.8B","title":"Qwen3.5 0.8B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=Qwen%2FQwen3.5-0.8B&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"Qwen/Qwen3.5-0.8B","hardware":"thor","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":303.9,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":135.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":27126,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":42053,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":367.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":141.5,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":46.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":48.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":51.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":56.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":54.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":54.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":33.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":35.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.19,"unit":"J","higherIsBetter":false,"digits":2,"baseline":0.42,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":5.3,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":2.38,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":57.4,"unit":"W","higherIsBetter":false,"digits":1,"baseline":56.8,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":62.4,"unit":"W","higherIsBetter":false,"digits":1,"baseline":60.3,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":21.7,"unit":"W","higherIsBetter":false,"digits":1,"baseline":24.8,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":3.56,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":3.93,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":2.99,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":1.98,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"8f86c75a8c4c6234","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.5-0.8B","hardware":"thor","variant":"huggingface_fp16_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":135.2,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":42053,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":141.5,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":48.8,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"GSM8K","value":56.7,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"IFEval","value":54.2,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"MMLU-Pro","value":35.4,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.42,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":2.38,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":56.8,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":60.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":24.8,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":3.93,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.5-0.8B","hardware":"thor","variant":"huggingface_int4_embedl-flashhead-mtp_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":303.9,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":135.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":27126,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":42053,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":367.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":141.5,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":46.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":48.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":51.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":56.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":54.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":54.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":33.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":35.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.19,"unit":"J","higherIsBetter":false,"digits":2,"baseline":0.42,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":5.3,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":2.38,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":57.4,"unit":"W","higherIsBetter":false,"digits":1,"baseline":56.8,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":62.4,"unit":"W","higherIsBetter":false,"digits":1,"baseline":60.3,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":21.7,"unit":"W","higherIsBetter":false,"digits":1,"baseline":24.8,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":3.56,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":3.93,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":2.99,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":1.98,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"8f86c75a8c4c6234","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.5-0.8B","hardware":"orin","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":178.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":15503,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":217.7,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":45.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":48.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":51.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":56.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":54.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":54.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":29.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":35.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.2,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":5.06,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":35.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":39.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":11.9,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":1.2,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":2.9,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":1.98,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"8c65d798668b3731","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.5-0.8B","hardware":"orin","variant":"huggingface_int4_embedl-flashhead-mtp_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":178.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":15503,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":217.7,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":45.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":48.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":51.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":56.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":54.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":54.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":29.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":35.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.2,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":5.06,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":35.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":39.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":11.9,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":1.2,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":2.9,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":1.98,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"8c65d798668b3731","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.5-0.8B","hardware":"orin-nx-16gb","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":81.8,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":6171,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":103.2,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":42.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":48.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":50.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":56.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":49.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":54.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":29.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":35.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.21,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":4.68,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":17.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":19.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":8.4,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":1.99,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":2.9,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":1.98,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.5-0.8B","hardware":"orin-nx-16gb","variant":"huggingface_int4_embedl-flashhead-mtp_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":81.8,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":6171,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":103.2,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":42.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":48.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":50.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":56.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":49.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":54.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":29.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":35.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.21,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":4.68,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":17.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":19.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":8.4,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":1.99,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":2.9,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":1.98,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.5-0.8B","hardware":"orin-nano-super","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":87.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":5695,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":115.5,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":44.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":48.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":51.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":56.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":50.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":54.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":32.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":35.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.21,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":4.84,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":18.1,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":19.4,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":7.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":2.81,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":2.9,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":1.98,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"bf949b847f10c3fd","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.5-0.8B","hardware":"orin-nano-super","variant":"huggingface_int4_embedl-flashhead-mtp_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":87.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":5695,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":115.5,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":44.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":48.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":51.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":56.7,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":50.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":54.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":32.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":35.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.21,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":4.84,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":18.1,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":19.4,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":7.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":2.81,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":2.9,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":1.98,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"bf949b847f10c3fd","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"Qwen/Qwen3.5-27B","title":"Qwen3.5 27B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=Qwen%2FQwen3.5-27B&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"Qwen/Qwen3.5-27B","hardware":"thor","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":23.4,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":941,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":38.6,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":88.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":88.9,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":96.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":96.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":88.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":88.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":81.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":82.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":3.58,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.28,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":83.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":100.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":23.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":24.22,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":21.3,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":20.29,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"2eda690e848feee4","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.5-27B","hardware":"thor","variant":"huggingface_int4_embedl-flashhead-mtp_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":23.4,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":941,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":38.6,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":88.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":88.9,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":96.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":96.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":88.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":88.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":81.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":82.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":3.58,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.28,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":83.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":100.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":23.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":24.22,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":21.3,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":20.29,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"2eda690e848feee4","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"Qwen/Qwen3.5-2B","title":"Qwen3.5 2B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=Qwen%2FQwen3.5-2B&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"Qwen/Qwen3.5-2B","hardware":"thor","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":108.9,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":60.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":13239,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":26771,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":124.4,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":62.3,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":63.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":66.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":77.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":80.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":65.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":47.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":52.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.58,"unit":"J","higherIsBetter":false,"digits":2,"baseline":0.99,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":1.74,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":1.01,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":62.7,"unit":"W","higherIsBetter":false,"digits":1,"baseline":59.9,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":69.7,"unit":"W","higherIsBetter":false,"digits":1,"baseline":68.3,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":22.6,"unit":"W","higherIsBetter":false,"digits":1,"baseline":22.2,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":4.72,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":7.1,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":4.24,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":3.23,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"c3a006ada8faa503","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.5-2B","hardware":"thor","variant":"huggingface_fp16_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":60.5,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":26771,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":62.3,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":66.1,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"GSM8K","value":80.4,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"IFEval","value":65.8,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"MMLU-Pro","value":52.3,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.99,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.01,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":59.9,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":68.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":22.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":7.1,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.5-2B","hardware":"thor","variant":"huggingface_int4_embedl_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":108.9,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":60.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":13239,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":26771,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":124.4,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":62.3,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":63.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":66.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":77.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":80.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":65.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":47.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":52.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.58,"unit":"J","higherIsBetter":false,"digits":2,"baseline":0.99,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":1.74,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":1.01,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":62.7,"unit":"W","higherIsBetter":false,"digits":1,"baseline":59.9,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":69.7,"unit":"W","higherIsBetter":false,"digits":1,"baseline":68.3,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":22.6,"unit":"W","higherIsBetter":false,"digits":1,"baseline":22.2,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":4.72,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":7.1,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":4.24,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":3.23,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"c3a006ada8faa503","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.5-2B","hardware":"orin","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":69.6,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":8047,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":80.2,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":64.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":66.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":77.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":80.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":64.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":50.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":52.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.58,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.71,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":40.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":47.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":1.2,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":4.15,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":3.23,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"4e1fc2ffec58dbd5","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.5-2B","hardware":"orin","variant":"huggingface_int4_embedl_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":69.6,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":8047,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":80.2,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":64.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":66.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":77.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":80.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":64.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":50.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":52.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.58,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.71,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":40.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":47.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":1.2,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":4.15,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":3.23,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"4e1fc2ffec58dbd5","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.5-2B","hardware":"orin-nx-16gb","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":34.9,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":3132,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":42.2,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":63.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":66.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":76.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":80.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":66,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":47.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":52.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.58,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.73,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":20.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":22.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":8.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":2.86,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":4.15,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":3.23,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.5-2B","hardware":"orin-nx-16gb","variant":"huggingface_int4_embedl_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":34.9,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":3132,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":42.2,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":63.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":66.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":76.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":80.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":66,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":47.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":52.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.58,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.73,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":20.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":22.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":8.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":2.86,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":4.15,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":3.23,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.5-2B","hardware":"orin-nano-super","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":36.2,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":2772,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":45.5,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":63.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":66.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":75.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":80.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":64.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":50.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":52.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.6,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.68,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":21.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":22,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":8,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":4.31,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":4.15,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":3.23,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"578c40016db12c26","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.5-2B","hardware":"orin-nano-super","variant":"huggingface_int4_embedl_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":36.2,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":2772,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":45.5,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":63.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":66.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":75.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":80.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":64.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":65.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":50.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":52.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.6,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.68,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":21.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":22,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":8,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":4.31,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":4.15,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":3.23,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"578c40016db12c26","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"Qwen/Qwen3.5-4B","title":"Qwen3.5 4B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=Qwen%2FQwen3.5-4B&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"Qwen/Qwen3.5-4B","hardware":"thor","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":65.6,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":27.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":5504,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":10917,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":80.5,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":28.4,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":81.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":82.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":93,"unit":"%","higherIsBetter":true,"digits":1,"baseline":93.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":79.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":81.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":71.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":72.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.12,"unit":"J","higherIsBetter":false,"digits":2,"baseline":2.39,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":0.89,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":0.42,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":73.8,"unit":"W","higherIsBetter":false,"digits":1,"baseline":65.7,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":84.6,"unit":"W","higherIsBetter":false,"digits":1,"baseline":84.7,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":23.1,"unit":"W","higherIsBetter":false,"digits":1,"baseline":22.7,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":9.7,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":11.99,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":7.45,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":6.44,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"97309249e59c181b","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.5-4B","hardware":"thor","variant":"huggingface_fp16_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":27.5,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":10917,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":28.4,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":82.3,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"GSM8K","value":93.3,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"IFEval","value":81.5,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"MMLU-Pro","value":72.1,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":2.39,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.42,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":65.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":84.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":22.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":11.99,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.5-4B","hardware":"thor","variant":"huggingface_int4_embedl-dflash_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":65.6,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":27.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":5504,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":10917,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":80.5,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":28.4,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":81.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":82.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":93,"unit":"%","higherIsBetter":true,"digits":1,"baseline":93.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":79.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":81.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":71.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":72.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.12,"unit":"J","higherIsBetter":false,"digits":2,"baseline":2.39,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":0.89,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":0.42,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":73.8,"unit":"W","higherIsBetter":false,"digits":1,"baseline":65.7,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":84.6,"unit":"W","higherIsBetter":false,"digits":1,"baseline":84.7,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":23.1,"unit":"W","higherIsBetter":false,"digits":1,"baseline":22.7,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":9.7,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":11.99,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":7.45,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":6.44,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"97309249e59c181b","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.5-4B","hardware":"orin","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":41.5,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":3235,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":51.8,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":80.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":82.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":92.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":93.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":79.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":81.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":69.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":72.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.12,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.9,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":46.4,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":54.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":5.77,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":7.37,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":6.45,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"fc4bec4e168a5bc4","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.5-4B","hardware":"orin","variant":"huggingface_int4_embedl-dflash_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":41.5,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":3235,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":51.8,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":80.6,"unit":"%","higherIsBetter":true,"digits":1,"baseline":82.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":92.7,"unit":"%","higherIsBetter":true,"digits":1,"baseline":93.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":79.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":81.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":69.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":72.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.12,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.9,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":46.4,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":54.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":5.77,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":7.37,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":6.45,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"fc4bec4e168a5bc4","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"Qwen/Qwen3.5-9B","title":"Qwen3.5 9B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=Qwen%2FQwen3.5-9B&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"Qwen/Qwen3.5-9B","hardware":"thor","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":38,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":15.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":3156,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":6390,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":46.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":15.5,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":83.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":85.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":94.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":95.6,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":83.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":83.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":73.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":76.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":2.07,"unit":"J","higherIsBetter":false,"digits":2,"baseline":4.51,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":0.48,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":0.22,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":78.6,"unit":"W","higherIsBetter":false,"digits":1,"baseline":67.9,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":93.4,"unit":"W","higherIsBetter":false,"digits":1,"baseline":99.5,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":22.7,"unit":"W","higherIsBetter":false,"digits":1,"baseline":24.8,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":12.46,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":20.06,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":11.06,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":10.05,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"afba33d403358399","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.5-9B","hardware":"thor","variant":"huggingface_fp16_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":15.1,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":6390,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":15.5,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":85.1,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"GSM8K","value":95.6,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"IFEval","value":83.4,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"MMLU-Pro","value":76.4,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":4.51,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.22,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":67.9,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":99.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":24.8,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":20.06,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.5-9B","hardware":"thor","variant":"huggingface_int4_embedl-mtp_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":38,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":15.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Prompt processing","value":3156,"unit":"tok/s","higherIsBetter":true,"digits":0,"baseline":6390,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Decode","value":46.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"baseline":15.5,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":83.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":85.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":94.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":95.6,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":83.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":83.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":73.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":76.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":2.07,"unit":"J","higherIsBetter":false,"digits":2,"baseline":4.51,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"Generated tokens per joule","value":0.48,"unit":"tok/J","higherIsBetter":true,"digits":2,"baseline":0.22,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, mean","value":78.6,"unit":"W","higherIsBetter":false,"digits":1,"baseline":67.9,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power while generating, peak","value":93.4,"unit":"W","higherIsBetter":false,"digits":1,"baseline":99.5,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Power at idle","value":22.7,"unit":"W","higherIsBetter":false,"digits":1,"baseline":24.8,"baselineLabel":"Baseline TensorRT Edge-LLM"},{"description":"Peak memory","value":12.46,"unit":"GiB","higherIsBetter":false,"digits":2,"baseline":20.06,"baselineLabel":"Baseline TensorRT Edge-LLM"}]},{"title":"Size","metrics":[{"description":"Total size","value":11.06,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":10.05,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"afba33d403358399","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.5-9B","hardware":"orin","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":25.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":1993,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":32.1,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":83.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":85.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":94.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":95.6,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":83.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":83.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":73.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":76.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.85,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.54,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":47.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":61.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":9.85,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":10.98,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":10.06,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"1be9ed7ed42e500c","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.5-9B","hardware":"orin","variant":"huggingface_int4_embedl-mtp_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":25.7,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":1993,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":32.1,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":83.9,"unit":"%","higherIsBetter":true,"digits":1,"baseline":85.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":94.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":95.6,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":83.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":83.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":73.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":76.4,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":1.85,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.54,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":47.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":61.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":12.6,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":9.85,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":10.98,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":10.06,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"1be9ed7ed42e500c","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"Qwen/Qwen3.6-27B","title":"Qwen3.6 27B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=Qwen%2FQwen3.6-27B&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"Qwen/Qwen3.6-27B","hardware":"thor","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":24.1,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":940,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":40.5,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":88.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":88.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":97.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":97,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":85.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":85.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":83.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":82.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":3.49,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.29,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":84,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":100.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":23.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":24.21,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":21.3,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":20.29,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"d7581840b7e5d6f5","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.6-27B","hardware":"thor","variant":"huggingface_int4_embedl-flashhead-mtp_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":24.1,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":940,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":40.5,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":88.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":88.3,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":97.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":97,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":85.8,"unit":"%","higherIsBetter":true,"digits":1,"baseline":85.2,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":83.2,"unit":"%","higherIsBetter":true,"digits":1,"baseline":82.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":3.49,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.29,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":84,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":100.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":23.2,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":24.21,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":21.3,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":20.29,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"d7581840b7e5d6f5","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"Qwen/Qwen3.6-35B-A3B","title":"Qwen3.6 35B-A3B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=Qwen%2FQwen3.6-35B-A3B&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"Qwen/Qwen3.6-35B-A3B","hardware":"thor","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":82.1,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":4366,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":116.5,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":85,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"GSM8K","value":95.6,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"IFEval","value":80.8,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"MMLU-Pro","value":78.6,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.81,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.23,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":66.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":77.1,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":22.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":24.58,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":22.63,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":21.62,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"b951edfb229597a1","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.6-35B-A3B","hardware":"thor","variant":"huggingface_int4_embedl-flashhead-mtp_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":82.1,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":4366,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":116.5,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":85,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"GSM8K","value":95.6,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"IFEval","value":80.8,"unit":"%","higherIsBetter":true,"digits":1,"headline":true},{"description":"MMLU-Pro","value":78.6,"unit":"%","higherIsBetter":true,"digits":1,"headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":0.81,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":1.23,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":66.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":77.1,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":22.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":24.58,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":22.63,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":21.62,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"b951edfb229597a1","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"Qwen/Qwen3.8-27B","title":"Qwen3.8 27B","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=Qwen%2FQwen3.8-27B&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"Qwen/Qwen3.8-27B","hardware":"thor","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":23.6,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":926,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":39.6,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":86.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":85.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":96.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":96.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":83.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":81.9,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":79.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":79.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":3.55,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.28,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":83.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":100.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":23.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":24.77,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":21.3,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":20.29,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"039c3dd747d129b6","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"Qwen/Qwen3.8-27B","hardware":"thor","variant":"huggingface_int4_embedl-flashhead-mtp_tensorrt-edgellm","metrics":[{"title":"Throughput","metrics":[{"description":"End-to-end","value":23.6,"unit":"tok/s","higherIsBetter":true,"digits":1,"headline":true},{"description":"Prompt processing","value":926,"unit":"tok/s","higherIsBetter":true,"digits":0},{"description":"Decode","value":39.6,"unit":"tok/s","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","metrics":[{"description":"Average of 3","value":86.3,"unit":"%","higherIsBetter":true,"digits":1,"baseline":85.8,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"GSM8K","value":96.4,"unit":"%","higherIsBetter":true,"digits":1,"baseline":96.1,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"IFEval","value":83.5,"unit":"%","higherIsBetter":true,"digits":1,"baseline":81.9,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true},{"description":"MMLU-Pro","value":79.1,"unit":"%","higherIsBetter":true,"digits":1,"baseline":79.5,"baselineLabel":"Baseline TensorRT Edge-LLM","headline":true}]},{"title":"Power and memory","metrics":[{"description":"Energy per generated token","value":3.55,"unit":"J","higherIsBetter":false,"digits":2,"headline":true},{"description":"Generated tokens per joule","value":0.28,"unit":"tok/J","higherIsBetter":true,"digits":2},{"description":"Power while generating, mean","value":83.7,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power while generating, peak","value":100.3,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Power at idle","value":23.5,"unit":"W","higherIsBetter":false,"digits":1},{"description":"Peak memory","value":24.77,"unit":"GiB","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Total size","value":21.3,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Container size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, on disk","value":1.74,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size","value":20.29,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Text","VLM","Generation"],"build":"039c3dd747d129b6","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/rdnet_base.nv_in1k","title":"RDNet Base","source":"timm/rdnet_base.nv_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Frdnet_base.nv_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/rdnet_base.nv_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.40537,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.70996,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":2.53772,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.82654,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":410.7,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":365.777,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.81,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"98034e297e87a30d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/rdnet_base.nv_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.73565,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":7.76379,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":128.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.81,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"635f9df686c6d6b8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/rdnet_base.nv_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":15.3544,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":15.387,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":64.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.81,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e42bb2388ac01e33","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/rdnet_base.nv_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.62317,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.6983,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":381,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.81,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"07db36cd08ee839d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/rdnet_base.nv_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.78968,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.81543,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":263.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.81,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2bbc2810639f4eaf","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/rdnet_large.nv_in1k","title":"RDNet Large","source":"timm/rdnet_large.nv_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Frdnet_large.nv_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/rdnet_large.nv_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.37503,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.13821,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":3.49316,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.29297,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":294.2,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":239.317,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.61,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.04,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d86da534607553e7","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/rdnet_large.nv_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":10.9426,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":10.9703,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":90.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.61,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.04,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8ea3fc7dbdf2b99d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/rdnet_large.nv_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":22.6984,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":22.7475,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":43.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.61,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.04,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2ce3eeac3f6cf22f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/rdnet_large.nv_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.95239,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.78762,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":4.02649,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.90765,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":252.9,"unit":"qps","digits":1,"baseline":208.801,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.61,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.04,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.62,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.1,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"8afb93ba46342af5","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/rdnet_large.nv_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.27575,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.30063,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":189.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.61,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.04,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"add097c1613cd358","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/rdnet_small.nv_in1k","title":"RDNet Small","source":"timm/rdnet_small.nv_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Frdnet_small.nv_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/rdnet_small.nv_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.97366,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.117,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":500.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.55,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"292ad709a24201a6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/rdnet_small.nv_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.1848,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.19702,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":160.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.55,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a79f89789fca0ccf","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/rdnet_small.nv_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":11.8953,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":11.9144,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":83.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.55,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"64f02d2b9334a82b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/rdnet_small.nv_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.09524,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.099,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":476.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.55,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.09,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.57,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"3c34d20b9a9a2432","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/rdnet_small.nv_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.1571,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.16608,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":316.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.55,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4c259fd837ed45b5","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/rdnet_tiny.nv_in1k","title":"RDNet Tiny","source":"timm/rdnet_tiny.nv_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Frdnet_tiny.nv_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/rdnet_tiny.nv_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.18415,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.33312,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1.30286,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.45334,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":830.5,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":737.534,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.61,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.12,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"08dc97e8cc7ee865","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/rdnet_tiny.nv_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.86299,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.87268,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":257.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.61,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.12,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ee60b7d9c1d7f747","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/rdnet_tiny.nv_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.43762,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":7.45352,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":133.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.61,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.12,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d452298f8584c622","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/rdnet_tiny.nv_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.12541,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.24709,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.12836,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.25623,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":887.4,"unit":"qps","digits":1,"baseline":800.786,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.61,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.12,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.79,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.27,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"220a3de12b4c0475","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/rdnet_tiny.nv_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.8977,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.90094,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":526.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.61,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.12,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2335e68f379012c5","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/regnetx_002.pycls_in1k","title":"RegNetX-002","source":"timm/regnetx_002.pycls_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnetx_002.pycls_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnetx_002.pycls_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.229175,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.276733,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3969.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":67.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":87.59,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"18a60c081da97f32","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_002.pycls_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.451471,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.455566,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2196.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":67.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":87.59,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a3534ce1278bc08f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_002.pycls_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.659186,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.663879,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1505.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":67.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":87.59,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"06f905957c5b5754","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_002.pycls_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.132605,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.156954,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.132874,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.157349,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":7458.4,"unit":"qps","digits":1,"baseline":6309.65,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":67.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":87.59,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.55,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":31.8,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"56e4a034dc3806f3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_002.pycls_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.171094,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.172058,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":5807.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":67.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":87.59,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a817e208b540c8dd","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/regnetx_004.pycls_in1k","title":"RegNetX-004","source":"timm/regnetx_004.pycls_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnetx_004.pycls_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnetx_004.pycls_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.316091,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.48291,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2891.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":71.65,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.4,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b28415563a978763","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_004.pycls_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.594373,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.598328,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1669.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":71.65,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.4,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"17eb4385faf07ac9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_004.pycls_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.852367,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.85675,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1165.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":71.65,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.4,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"cdfd98135f61f656","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_004.pycls_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.195345,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.23639,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.195557,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.236816,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":5078.1,"unit":"qps","digits":1,"baseline":4199.57,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":71.65,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.4,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.58,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":60.7,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"d7d93278de644c4a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_004.pycls_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.398957,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.400635,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2499.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":71.65,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.4,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"062b788762d2f441","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/regnetx_004_tv.tv2_in1k","title":"RegNetX-004 (tv)","source":"timm/regnetx_004_tv.tv2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnetx_004_tv.tv2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnetx_004_tv.tv2_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.300907,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.373169,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3071.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":73.06,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.23,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d56836cc2d6f0c4c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_004_tv.tv2_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.713696,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.718323,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1391.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":73.06,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.23,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5f143dd3c3b76ac6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_004_tv.tv2_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.03565,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.04065,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":959.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":73.06,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.23,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"dc0810bde215170c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_004_tv.tv2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.253358,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.253723,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3926.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":73.06,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.23,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.58,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":64.4,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"fa484bd24993b4b9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_004_tv.tv2_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.474636,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.47821,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2102.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":73.06,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.23,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1523265bdf2a14b4","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/regnetx_006.pycls_in1k","title":"RegNetX-006","source":"timm/regnetx_006.pycls_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnetx_006.pycls_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnetx_006.pycls_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.92513,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.06195,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":511.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":73.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.34,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4671a09687d48f63","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_006.pycls_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.421061,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.744021,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.421997,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.744507,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":2366.1,"unit":"qps","digits":1,"baseline":1341.69,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":73.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.34,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":75.3,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"315fd85dbf805088","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_006.pycls_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.583025,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.585754,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1711.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":73.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.34,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9c5dcbc8a86b60a8","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/regnetx_008.pycls_in1k","title":"RegNetX-008","source":"timm/regnetx_008.pycls_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnetx_008.pycls_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnetx_008.pycls_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.238407,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.297241,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3807,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.43,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.99,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f22e31fdb43e3e56","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_008.pycls_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.536075,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.540924,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1851.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.43,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.99,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"56bbf80e328ced22","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_008.pycls_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.865235,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.869263,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1147.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.43,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.99,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"71e798614f6c2321","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_008.pycls_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.166118,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.210592,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.166504,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.211548,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":5965.9,"unit":"qps","digits":1,"baseline":4718.11,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":74.43,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.99,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.6,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":84.6,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"94ceb6608ba42270","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_008.pycls_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.324522,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.325684,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3071.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.43,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.99,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"cedba2913fc9cf71","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/regnetx_016.pycls_in1k","title":"RegNetX-016","source":"timm/regnetx_016.pycls_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnetx_016.pycls_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnetx_016.pycls_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.541489,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.01193,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.543518,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.013,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1841.8,"unit":"qps","digits":1,"baseline":986.845,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.64,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.11,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.63,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.11,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"0c2df0d720e9f766","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_016.pycls_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.758879,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.760437,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1315.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.64,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.11,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"820fc20421cababd","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/regnetx_032.pycls_in1k","title":"RegNetX-032","source":"timm/regnetx_032.pycls_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnetx_032.pycls_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnetx_032.pycls_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.49694,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.5368,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":655.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.82,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3f12a8a531de575a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_032.pycls_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.41709,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.42328,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":411.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.82,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"916f60a2e5e73a78","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_032.pycls_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.60837,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.61639,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":275.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.82,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c2c966b51112d5bb","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_032.pycls_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.992787,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.15532,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.99469,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.15747,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1005.7,"unit":"qps","digits":1,"baseline":864.539,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.82,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.71,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.18,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"c2645889ffa18266","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_032.pycls_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.30238,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.30457,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":767.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.82,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5bf1fbdca9affa89","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/regnetx_040.pycls_in1k","title":"RegNetX-040","source":"timm/regnetx_040.pycls_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnetx_040.pycls_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnetx_040.pycls_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.61671,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.68384,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":378,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.06,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.49,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a1bfd65ffce2c692","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_040.pycls_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.08948,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.61788,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.09186,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.62213,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":916.6,"unit":"qps","digits":1,"baseline":617.453,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.06,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.49,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.78,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.26,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"3fe8c6ae3a991b85","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_040.pycls_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.29869,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.3006,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":769.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.06,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.49,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a2711fcd2ea6169a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/regnetx_064.pycls_in1k","title":"RegNetX-064","source":"timm/regnetx_064.pycls_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnetx_064.pycls_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnetx_064.pycls_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.78038,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.81879,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":552.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.75,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.31,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"03cd4ecd75cbe418","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_064.pycls_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.17796,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.46575,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.18103,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.47009,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":847.6,"unit":"qps","digits":1,"baseline":681.502,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.75,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.31,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.83,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.31,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"dde0bcdb72f101eb","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_064.pycls_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.22827,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.23291,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":813.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.75,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.31,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6e3ad1e5b4cc5c4c","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/regnetx_080.pycls_in1k","title":"RegNetX-080","source":"timm/regnetx_080.pycls_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnetx_080.pycls_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnetx_080.pycls_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.11091,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.15723,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":881,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.98,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.4,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"da9dd0280b98a4cd","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_080.pycls_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.77811,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.78571,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":562,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.98,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.4,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.48,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"966787effb9d723a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_080.pycls_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.74415,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.75,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":573,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.98,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.4,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"21b9b48759917313","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/regnet_x_1_6gf","title":"RegNetX-1.6GF","source":"torchvision/regnet_x_1_6gf","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fregnet_x_1_6gf&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/regnet_x_1_6gf","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.7542248163372278,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":1325.8646206529506,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":77.724,"unit":"%","digits":2,"baseline":79.668,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":93.824,"unit":"%","digits":2,"baseline":94.922,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/regnetx_120.pycls_in1k","title":"RegNetX-120","source":"timm/regnetx_120.pycls_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnetx_120.pycls_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnetx_120.pycls_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.05072,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.17152,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1.09271,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.24054,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":947,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":846.362,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.52,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5a3f3cbddb7d1d9b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_120.pycls_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.48149,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.4881,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":400.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.52,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"61e1d477574088a2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_120.pycls_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.54482,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.5542,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":218.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.52,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5d375d30c4336d2a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_120.pycls_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.2603,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.35809,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":442.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.52,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.08,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.56,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"0486ca2c7db9b188","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_120.pycls_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.10527,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.11469,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":474.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.52,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c3742af02abfcc18","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/regnetx_160.pycls_in1k","title":"RegNetX-160","source":"timm/regnetx_160.pycls_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnetx_160.pycls_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnetx_160.pycls_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.735019,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.09708,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.791138,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.17267,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1315,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":890.033,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.72,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.78,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7352899b4d65b2fc","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_160.pycls_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.37619,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.38116,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":722.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.72,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.78,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"cacfaa61c46972dc","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_160.pycls_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.95139,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.95782,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":336.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.72,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.78,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2182c7bb912bda42","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_160.pycls_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.629046,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.05087,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.631165,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.05359,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1585.6,"unit":"qps","digits":1,"baseline":950.181,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.72,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.78,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.13,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.61,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"9d60047cab2e405d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_160.pycls_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.849695,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.85199,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1175.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.72,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.78,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8b9cc2cddaa92e09","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/regnet_x_16gf","title":"RegNetX-16GF","source":"torchvision/regnet_x_16gf","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fregnet_x_16gf&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/regnet_x_16gf","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.7172888051718473,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":1394.1385851692207,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":80.44,"unit":"%","digits":2,"baseline":82.716,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":95.132,"unit":"%","digits":2,"baseline":96.196,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"torchvision/regnet_x_3_2gf","title":"RegNetX-3.2GF","source":"torchvision/regnet_x_3_2gf","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fregnet_x_3_2gf&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/regnet_x_3_2gf","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.2411531992256641,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":805.7023102578185,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":79.81200000000001,"unit":"%","digits":2,"baseline":81.196,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":94.81,"unit":"%","digits":2,"baseline":95.43,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/regnetx_320.pycls_in1k","title":"RegNetX-320","source":"timm/regnetx_320.pycls_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnetx_320.pycls_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnetx_320.pycls_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.03718,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.39232,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":2.15997,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.51233,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":487.7,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":413.932,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.85,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"90c8297d9e318896","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_320.pycls_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.72305,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.86523,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":211.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.85,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.85,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.33,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e622cf169242e15d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnetx_320.pycls_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.6574,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.76001,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":214.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.85,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ab4fcca88692bdf1","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/regnet_x_32gf","title":"RegNetX-32GF","source":"torchvision/regnet_x_32gf","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fregnet_x_32gf&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/regnet_x_32gf","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":4.617102780030109,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":216.5860383973256,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":81.158,"unit":"%","digits":2,"baseline":83.014,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":95.39800000000001,"unit":"%","digits":2,"baseline":96.288,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"torchvision/regnet_x_400mf","title":"RegNetX-400MF","source":"torchvision/regnet_x_400mf","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnetx_004_tv.tv2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/regnet_x_400mf","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.49178377987118443,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":2033.4139533067466,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":72.676,"unit":"%","digits":2,"baseline":74.864,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":91.2,"unit":"%","digits":2,"baseline":92.322,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"torchvision/regnet_x_800mf","title":"RegNetX-800MF","source":"torchvision/regnet_x_800mf","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fregnet_x_800mf&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/regnet_x_800mf","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.3220563568174839,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":3105.046613213479,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":75.882,"unit":"%","digits":2,"baseline":77.522,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":92.86,"unit":"%","digits":2,"baseline":93.826,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"torchvision/regnet_x_8gf","title":"RegNetX-8GF","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fregnet_x_8gf&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/regnet_x_8gf","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":2.136,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":2.213,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":467.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":80.13,"unit":"%","headline":true},{"description":"Top-5","value":94.97,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/regnet_x_8gf","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":2.136,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":2.213,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":467.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":80.13,"unit":"%","headline":true},{"description":"Top-5","value":94.97,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/regnety_002.pycls_in1k","title":"RegNetY-002","source":"timm/regnety_002.pycls_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnety_002.pycls_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnety_002.pycls_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.447547,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.481018,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2103.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":68.97,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":88.75,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8ed5c87e7220cdfb","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_002.pycls_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.780571,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.783813,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1272.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":68.97,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":88.75,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"71fa52575a7392cb","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_002.pycls_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.12238,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.12872,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":885.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":68.97,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":88.75,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"bad1429533671e8e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_002.pycls_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.309715,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.309998,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3215.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":68.97,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":88.75,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.56,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":37.4,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"af5715e8c0b8bc77","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_002.pycls_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.441914,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.44342,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2257.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":68.97,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":88.75,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"fb62ba9b5912ffcb","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/regnety_004.pycls_in1k","title":"RegNetY-004","source":"timm/regnety_004.pycls_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnety_004.pycls_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnety_004.pycls_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.480597,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.548035,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1974.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.82,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3634d7af5721b282","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_004.pycls_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.987611,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.992249,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1006,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.82,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7aafd86c0ee0df50","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_004.pycls_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.51348,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.51953,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":656.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.82,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"724cdc27199ce3c7","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_004.pycls_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.400963,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.40155,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2485.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":72.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.82,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.57,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":51.2,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"d8038c8013cb7d5a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_004.pycls_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.632861,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.634705,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1577.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.82,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7000c8ad22391b5c","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/regnety_006.pycls_in1k","title":"RegNetY-006","source":"timm/regnety_006.pycls_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnety_006.pycls_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnety_006.pycls_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.554,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.555511,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1800.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":74.73,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.31,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.6,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.08,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"36b059ef734998e0","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/regnety_008.pycls_in1k","title":"RegNetY-008","source":"timm/regnety_008.pycls_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnety_008.pycls_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnety_008.pycls_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.42418,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.507812,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2230.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.33,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7a77dc5a9295697c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_008.pycls_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.880383,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.885223,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1128.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.33,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"bee30d5f0b96680d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_008.pycls_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.29025,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.29498,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":770.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.33,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ac716daa8b1baa0a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_008.pycls_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.366978,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.367615,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2713.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":75.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.33,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":73.2,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"9d7b856eb5c4bb0f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_008.pycls_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.638,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.640076,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1564.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.33,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ae1c639954710605","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/regnety_008_tv.tv2_in1k","title":"RegNetY-008 (tv)","source":"timm/regnety_008_tv.tv2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnety_008_tv.tv2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnety_008_tv.tv2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.432899,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.433777,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2303.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":0.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":2.91,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.07,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"4d887bc2901a9e12","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/regnety_016.pycls_in1k","title":"RegNetY-016","source":"timm/regnety_016.pycls_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnety_016.pycls_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnety_016.pycls_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.40203,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.4762,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":291.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.41,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.09,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"27e6bfed03e8a215","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_016.pycls_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.17616,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.65598,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.17743,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.65704,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":849.1,"unit":"qps","digits":1,"baseline":603.363,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.41,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.09,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.65,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.13,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"c9c25922b44b2844","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_016.pycls_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.66039,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.6698,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":601.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.41,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.09,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7c9b01694de7bf5e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/regnety_032.pycls_in1k","title":"RegNetY-032","source":"timm/regnety_032.pycls_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnety_032.pycls_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnety_032.pycls_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.58267,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.67273,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":276.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.97,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.93,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"83548455d213b86a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_032.pycls_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.09176,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.55358,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.09467,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.55566,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":914.6,"unit":"qps","digits":1,"baseline":643.112,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.97,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.93,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.75,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.23,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"07061f7b88097444","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_032.pycls_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.51251,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.51581,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":660.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.97,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.93,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"51d72e22ac716f68","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/regnety_040.pycls_in1k","title":"RegNetY-040","source":"timm/regnety_040.pycls_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnety_040.pycls_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnety_040.pycls_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.733263,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.858706,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.8396,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.963013,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1349.3,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1136.37,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.96,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.95,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f17558071243fe4a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_040.pycls_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.48146,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.48645,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":671,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.96,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.95,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"de55b140705017ca","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_040.pycls_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.38406,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.39105,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":417.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.96,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.95,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"435fa2b858cf90c5","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_040.pycls_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.654768,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.655762,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1524.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.96,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.95,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.75,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.23,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"02f311b0c6267c7d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_040.pycls_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.07104,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.07318,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":932.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.96,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.95,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a53855c3119bdcfe","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/regnety_064.pycls_in1k","title":"RegNetY-064","source":"timm/regnety_064.pycls_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnety_064.pycls_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnety_064.pycls_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.13933,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.19781,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":461.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.3,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.1,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2b64828b4dbeb3ff","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_064.pycls_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.27413,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.30188,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":439.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.3,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.1,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.89,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.37,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"c15a2b5f434352f0","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_064.pycls_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.57045,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.58246,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":388.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.3,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.1,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"38bca7dc468fbe8c","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/regnety_080.pycls_in1k","title":"RegNetY-080","source":"timm/regnety_080.pycls_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnety_080.pycls_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnety_080.pycls_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.14438,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.1983,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":460.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.11,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"baebdb97a77d4535","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_080.pycls_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.76215,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.76825,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":567,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.11,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.98,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.46,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"a2deb74fd0277a90","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_080.pycls_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.79507,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.80231,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":556.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.11,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2b368b87642836f9","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/regnety_080_tv.tv2_in1k","title":"RegNetY-080 (tv)","source":"timm/regnety_080_tv.tv2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnety_080_tv.tv2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnety_080_tv.tv2_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.21007,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.25171,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":446.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.13,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.02,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"afa13cd61d7abd58","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_080_tv.tv2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.80404,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.81604,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":553.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.13,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.02,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.98,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.46,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"2903f77e86e71173","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_080_tv.tv2_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.91621,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.91937,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":521.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.13,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.02,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"288a2518a15940f3","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/regnet_y_1_6gf","title":"RegNetY-1.6GF","source":"torchvision/regnet_y_1_6gf","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fregnet_y_1_6gf&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/regnet_y_1_6gf","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.6395716927945614,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":609.9153848500239,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":79.896,"unit":"%","digits":2,"baseline":80.876,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":94.93,"unit":"%","digits":2,"baseline":95.444,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/regnety_120.pycls_in1k","title":"RegNetY-120","source":"timm/regnety_120.pycls_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnety_120.pycls_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnety_120.pycls_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.39795,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.50281,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":713.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.22,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.57,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"24fc3d88af779708","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_120.pycls_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.64511,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.65314,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":272.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.22,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.57,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3f477870f0f6a39e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_120.pycls_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.39586,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.40674,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":155.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.22,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.57,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"fb1c8d85f590e220","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_120.pycls_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.68233,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.78632,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":372.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.22,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.57,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.14,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.62,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"bdf38101f86fdf79","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_120.pycls_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.60404,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.62823,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":383.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.22,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.57,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"10824ab900a3efa0","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/regnety_1280.seer","title":"RegNetY-1280 (SEER)","source":"timm/regnety_1280.seer","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnety_1280.seer&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnety_1280.seer","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":16.367,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":16.8722,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":61.1,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":6.85,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":6.33,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"65f9f22c21360388","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/regnet_y_128gf","title":"RegNetY-128GF","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fregnet_y_128gf&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/regnet_y_128gf","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":18.036,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":18.89,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":55.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":82.56,"unit":"%","headline":true},{"description":"Top-5","value":96.87,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/regnet_y_128gf","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":18.036,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":18.89,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":55.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":82.56,"unit":"%","headline":true},{"description":"Top-5","value":96.87,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/regnety_160.deit_in1k","title":"RegNetY-160","source":"timm/regnety_160.deit_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnety_160.deit_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnety_160.deit_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.62828,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.01124,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1.73816,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.11597,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":612.6,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":493.53,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.27,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"026872b662ab1f32","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_160.deit_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.2086,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.22546,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":236.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.27,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5371707c0a79e721","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_160.deit_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.98781,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":8.04089,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":124.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.27,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"153820e7dda95ad7","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_160.deit_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.84461,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.91339,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":351.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.27,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.98,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"a3ff8dc1f6806f37","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_160.deit_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.93155,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.94623,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":341,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.27,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"498701eab0c2e877","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/regnet_y_16gf","title":"RegNetY-16GF","source":"torchvision/regnet_y_16gf","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fregnet_y_16gf&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/regnet_y_16gf","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":3.011354934424162,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":332.07643129959445,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":81.08,"unit":"%","digits":2,"baseline":82.886,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":95.336,"unit":"%","digits":2,"baseline":96.328,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/regnety_2560.seer_ft_in1k","title":"RegNetY-2560 (SEER)","source":"timm/regnety_2560.seer_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnety_2560.seer_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnety_2560.seer_ft_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":27.4682,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":27.9658,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":36.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.37,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"92a6c05ced657092","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_2560.seer_ft_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":189.774,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":190.447,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":5.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.37,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"dbaa22c04ba2338d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_2560.seer_ft_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":496.551,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":496.957,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":85.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.37,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1adb114f541206f7","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_2560.seer_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":98.7569,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":101.468,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":10.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":85.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.37,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":16.03,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":15.51,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"2dcd59e743e78401","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/regnet_y_3_2gf","title":"RegNetY-3.2GF","source":"torchvision/regnet_y_3_2gf","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fregnet_y_3_2gf&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/regnet_y_3_2gf","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.4914803020656109,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":670.4748286752831,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":80.616,"unit":"%","digits":2,"baseline":81.982,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":95.28999999999999,"unit":"%","digits":2,"baseline":95.972,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/regnety_320.pycls_in1k","title":"RegNetY-320","source":"timm/regnety_320.pycls_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnety_320.pycls_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnety_320.pycls_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.27673,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.63776,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":2.38928,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.75085,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":436.3,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":374.478,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.68,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1ae04a2f1c3fb688","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_320.pycls_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.56419,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.71661,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":179.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.68,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.27,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.75,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"c71728c9c6243f6d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/regnety_320.pycls_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.86774,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.98267,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":205.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.68,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6e7a9d02a5c8a7c2","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/regnety_320.seer","title":"RegNetY-320 (SEER)","source":"timm/regnety_320.seer","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnety_320.seer&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnety_320.seer","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.3553,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.49084,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":186.7,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.07,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.55,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"057427d38e7b9506","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/regnet_y_32gf","title":"RegNetY-32GF","source":"torchvision/regnet_y_32gf","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fregnet_y_32gf&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/regnet_y_32gf","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":5.782056041061878,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":172.9488598689454,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":82.806,"unit":"%","digits":2,"baseline":83.368,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":96.316,"unit":"%","digits":2,"baseline":96.498,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"torchvision/regnet_y_400mf","title":"RegNetY-400MF","source":"torchvision/regnet_y_400mf","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fregnet_y_400mf&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/regnet_y_400mf","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.950058177113533,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":1052.5671207190703,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":73.988,"unit":"%","digits":2,"baseline":75.804,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":91.788,"unit":"%","digits":2,"baseline":92.742,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/regnety_640.seer","title":"RegNetY-640 (SEER)","source":"timm/regnety_640.seer","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnety_640.seer&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnety_640.seer","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.58477,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":9.90454,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":104.3,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.56,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.04,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"4bb6e2d802d6b688","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/regnet_y_800mf","title":"RegNetY-800MF","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fregnet_y_800mf&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/regnet_y_800mf","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":657.379,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":658.386,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":1518.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.13,"unit":"%","headline":true},{"description":"Top-5","value":0.54,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/regnet_y_800mf","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":657.379,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":658.386,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":1518.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.13,"unit":"%","headline":true},{"description":"Top-5","value":0.54,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/regnet_y_8gf","title":"RegNetY-8GF","source":"torchvision/regnet_y_8gf","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fregnet_y_8gf&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/regnet_y_8gf","hardware":"thor","precision":"FP8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":2.29517,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.96802,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Throughput","value":441.314,"unit":"qps","digits":1,"baseline":509.791,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNette validation · 3,925 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":90.80254777070064,"unit":"%","digits":2,"baseline":90.70063694267516,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Top-5","value":99.51592356687898,"unit":"%","digits":2,"baseline":99.56687898089172,"baselineLabel":"Baseline TensorRT"}]},{"title":"Memory","metrics":[{"description":"Peak device memory","value":62,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":94,"baselineLabel":"Baseline TensorRT","improvementWord":"smaller","headline":true}]},{"title":"Size","metrics":[{"description":"Engine size","value":46.6,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":77.1,"baselineLabel":"Baseline TensorRT","improvementWord":"smaller","headline":true}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"torchvision/regnet_y_8gf","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":2.0295645017176867,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":492.71654049608543,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":82.224,"unit":"%","digits":2,"baseline":82.828,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":96.044,"unit":"%","digits":2,"baseline":96.33,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/regnetz_c16_evos.ch_in1k","title":"RegNetZ-C16 (evos)","source":"timm/regnetz_c16_evos.ch_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fregnetz_c16_evos.ch_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/regnetz_c16_evos.ch_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":88.3361,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":88.3398,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":11.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.25,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.91,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.68,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.16,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"c1c47b3e53d24d9b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/repghostnet_050.in1k","title":"RepGhostNet 0.5×","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Frepghostnet_050.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/repghostnet_050.in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":798.706,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":799.5,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":1249.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":56.25,"unit":"%","headline":true},{"description":"Top-5","value":79.66,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/repghostnet_058.in1k","title":"RepGhostNet 0.58×","source":"timm/repghostnet_058.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Frepghostnet_058.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/repghostnet_058.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.778948,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.779724,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1281.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":67.39,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":87.34,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.55,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.03,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"2625ed8e2fcb256d","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/repghostnet_080.in1k","title":"RepGhostNet 0.8×","source":"timm/repghostnet_080.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Frepghostnet_080.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/repghostnet_080.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.513456,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.514282,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1941.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":71.06,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.84,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.56,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.04,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"01717f5ded29b9e2","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/repghostnet_100.in1k","title":"RepGhostNet 1.0×","source":"timm/repghostnet_100.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Frepghostnet_100.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/repghostnet_100.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.532797,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.533325,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1871.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":73.08,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.97,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.57,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.05,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"a636661bd344e911","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/repghostnet_111.in1k","title":"RepGhostNet 1.11×","source":"timm/repghostnet_111.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Frepghostnet_111.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/repghostnet_111.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.939593,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.940491,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1062.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":71.22,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.02,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.57,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":54,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"82bdf85f30001d8d","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/repghostnet_130.in1k","title":"RepGhostNet 1.3×","source":"timm/repghostnet_130.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Frepghostnet_130.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/repghostnet_130.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.717261,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.814941,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1384.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.5,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.42,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"070539ec31968844","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/repghostnet_130.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.27952,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.29169,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":303.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.5,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.42,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"85f0e45e8b9dc0b8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/repghostnet_130.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.58671,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.60266,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":178.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.5,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.42,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6e4a6a06bbaab6b6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/repghostnet_130.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.93578,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.93689,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1067,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":75.5,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.42,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.58,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":64.9,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"e10428838cbf268d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/repghostnet_130.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.37222,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.38556,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":728.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.5,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.42,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8a1aa7fec60d98c1","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/repghostnet_150.in1k","title":"RepGhostNet 1.5×","source":"timm/repghostnet_150.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Frepghostnet_150.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/repghostnet_150.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.724614,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.832458,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1370.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.79,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3186143ad34c2350","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/repghostnet_150.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.37611,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.38849,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":294.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.79,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a75d17e954db5398","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/repghostnet_150.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.13695,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.14929,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":162.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.79,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"645ae8bfed11e6af","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/repghostnet_150.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.992419,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.993469,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1006.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.79,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.6,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":77.6,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"bf407a0a8dd4d4a3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/repghostnet_150.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.44869,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.45203,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":689.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.79,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"dab21443f4d8b854","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/repghostnet_200.in1k","title":"RepGhostNet 2.0×","source":"timm/repghostnet_200.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Frepghostnet_200.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/repghostnet_200.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.826886,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.931091,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1195,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.87,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"218837fdeba5aa69","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/repghostnet_200.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.77944,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.79297,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":263.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.87,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"79aa345e36f0f78a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/repghostnet_200.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.90434,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.9173,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":144.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.87,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b5611476d70aaec9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/repghostnet_200.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.08111,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.08295,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":923.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.87,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.63,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.11,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e139242c81749522","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/repghostnet_200.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.57074,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.57312,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":636.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.87,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4c8b7ca42a72fc09","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/repvit_m1.dist_in1k","title":"RepViT-M1","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Frepvit_m1.dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/repvit_m1.dist_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":978.271,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":979.431,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":1020.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":44.98,"unit":"%","headline":true},{"description":"Top-5","value":68.88,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/repvit_m1_5.dist_300e_in1k","title":"RepViT-M1.5","source":"timm/repvit_m1_5.dist_300e_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Frepvit_m1_5.dist_300e_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/repvit_m1_5.dist_300e_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.65053,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.66766,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":605.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.66,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.82,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.71,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.19,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"1f74f2c66e9ce825","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/repvit_m2.dist_in1k","title":"RepViT-M2","source":"timm/repvit_m2.dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Frepvit_m2.dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/repvit_m2.dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.26799,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.27136,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":784.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.59,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.63,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.11,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"944763923929e426","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/repvit_m3.dist_in1k","title":"RepViT-M3","source":"timm/repvit_m3.dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Frepvit_m3.dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/repvit_m3.dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.98196,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.983704,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1016.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.17,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.5,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.64,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.12,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"1b6958e4621f6f67","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/res2net101_26w_4s.in1k","title":"Res2Net-101 (26w, 4s)","source":"timm/res2net101_26w_4s.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fres2net101_26w_4s.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/res2net101_26w_4s.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.52066,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.66583,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":655.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4705e5c7b744a3b0","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/res2net101_26w_4s.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.10258,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.10571,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":475.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.03,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.51,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"a072ed5be86c6952","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/res2net101_26w_4s.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.73102,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.73901,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":267.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4f1fa458d9dc2099","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/res2net101d.in1k","title":"Res2Net-101-D","source":"timm/res2net101d.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fres2net101d.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/res2net101d.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.4941,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.66144,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":666.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.64,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.03,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a9b671e1eebbfd9a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/res2net101d.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.52947,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.55035,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":152.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.64,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.03,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"63ac159c91f8bc67","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/res2net101d.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.26702,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.27002,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":440.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.64,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.03,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.03,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.51,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"48cea8560b53149c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/res2net101d.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.71295,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.7168,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":269.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.64,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.03,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"edc2053300390d48","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/res2net50_14w_8s.in1k","title":"Res2Net-50 (14w, 8s)","source":"timm/res2net50_14w_8s.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fres2net50_14w_8s.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/res2net50_14w_8s.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.26829,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.33759,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":772.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.58,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.53,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"29e6b722dc3406de","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/res2net50_14w_8s.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.9825,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.99921,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":166.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.58,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.53,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"65c644b25018f60b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/res2net50_14w_8s.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.78994,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.79236,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":558.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.58,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.53,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.8,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.28,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"bcc4a7b19bfbfe16","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/res2net50_14w_8s.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.89784,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.91827,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":345,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.58,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.53,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"84500bf14b70f396","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/res2net50_26w_4s.in1k","title":"Res2Net-50 (26w, 4s)","source":"timm/res2net50_26w_4s.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fres2net50_26w_4s.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/res2net50_26w_4s.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.782289,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.931074,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.891357,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.02948,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1272.7,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1056.4,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.58,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"80d724f9d51f9aa3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/res2net50_26w_4s.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.38688,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.4007,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":293.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.58,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"dc4a93fbced3d350","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/res2net50_26w_4s.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.97132,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.972351,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1027.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.58,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.81,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.29,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"c737b365119a2a3f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/res2net50_26w_4s.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.79225,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.79718,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":557.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.58,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d2f91519eb3a44d7","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/res2net50_26w_6s.in1k","title":"Res2Net-50 (26w, 6s)","source":"timm/res2net50_26w_6s.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fres2net50_26w_6s.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/res2net50_26w_6s.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.10798,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.24084,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":898.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.86,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ecf8dde8f60beb49","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/res2net50_26w_6s.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.37457,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.38867,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":185.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.86,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a0164adeee46032e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/res2net50_26w_6s.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.70226,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.70532,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":586.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.86,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.94,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.42,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e613a07f4208e34a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/res2net50_26w_6s.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.7502,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.75446,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":363.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.86,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"fcca58f27655922b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/res2net50_48w_2s.in1k","title":"Res2Net-50 (48w, 2s)","source":"timm/res2net50_48w_2s.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fres2net50_48w_2s.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/res2net50_48w_2s.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.428956,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.499695,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2292.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.12,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.34,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4a2c71de871204e9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/res2net50_48w_2s.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.65548,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.66199,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":600.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.12,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.34,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5fc77db645b97683","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/res2net50_48w_2s.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.89569,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.90686,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":343.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.12,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.34,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4351a2e4828ef03a","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/res2net50_48w_2s.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.467417,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.468445,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2131.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.12,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.34,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.81,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.28,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"280a091c75881efb","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/res2net50_48w_2s.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.907123,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.909363,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1101.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.12,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.34,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e36615beec20f810","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/res2net50d.in1k","title":"Res2Net-50-D","source":"timm/res2net50d.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fres2net50d.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/res2net50d.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.824517,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.9151,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1206.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1165d97a146668ff","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/res2net50d.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.6525,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.6665,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":272.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5730ca9e242c45d7","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/res2net50d.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.06088,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.06335,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":941,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.81,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.29,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"624d811558e7c8d6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/res2net50d.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.87194,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.87555,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":533.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2943cd3f8a26c78e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/res2next50.in1k","title":"Res2NeXt-50","source":"timm/res2next50.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fres2next50.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/res2next50.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.616694,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.697876,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1608.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.59,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.54,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7675c6681254f808","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/res2next50.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.96758,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.97772,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":335.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.59,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.54,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"201cb38af294ce04","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/res2next50.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.7819,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.79114,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":208,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.59,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.54,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1d912eaba963d93f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/res2next50.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.04729,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.04907,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":953.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.59,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.54,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.8,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.28,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e73106da86eaf6fd","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/res2next50.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.51047,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.51483,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":398.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.59,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.54,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"40bf696b651194ca","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resmlp_big_24_224.fb_distilled_in1k","title":"ResMLP Big-24 (224)","source":"timm/resmlp_big_24_224.fb_distilled_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresmlp_big_24_224.fb_distilled_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resmlp_big_24_224.fb_distilled_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.09829,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.15631,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":320.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.58,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.64,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"444cd3163d14ea26","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resmlp_big_24_224.fb_distilled_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.72224,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":8.74866,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":114.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.58,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.64,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"738a9d50bb01917b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resmlp_big_24_224.fb_distilled_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":23.775,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":23.8569,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":41.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.58,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.64,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2df13a2df4d7b101","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resmlp_big_24_224.fb_distilled_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":10.7466,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":11.2856,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":93,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.58,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.64,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.31,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.79,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"5279fc8788a7e5d8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resmlp_big_24_224.fb_distilled_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":20.6157,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":20.9797,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":48.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.58,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.64,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5dbef1f76e8e8d1a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resmlp_12_224.fb_dino","title":"ResMLP-12 (224, DINO)","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresmlp_12_224.fb_dino&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resmlp_12_224.fb_dino","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":700.134,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":702.209,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":1425.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.13,"unit":"%","headline":true},{"description":"Top-5","value":0.62,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resmlp_36_224.fb_distilled_in1k","title":"ResMLP-36 (224)","source":"timm/resmlp_36_224.fb_distilled_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresmlp_36_224.fb_distilled_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resmlp_36_224.fb_distilled_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.35265,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.47186,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":725,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.18,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.5,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"288bf062a36e7f26","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resmlp_36_224.fb_distilled_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.52972,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.54175,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":393.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.18,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.5,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"91733c4dfc314a79","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resmlp_36_224.fb_distilled_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.08569,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.10339,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":195.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.18,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.5,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3e7008ea76621727","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resmlp_36_224.fb_distilled_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.95337,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.02496,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":511.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.18,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.5,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.14,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.62,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"d14e6dadc6413aca","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resmlp_36_224.fb_distilled_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.33376,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.35669,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":299.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.18,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.5,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"51d27f205179bd79","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnest101e.in1k","title":"ResNeSt-101-E","source":"timm/resnest101e.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnest101e.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnest101e.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.30355,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.43634,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":430.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.89,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6ded361cf526ca41","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnest101e.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.68679,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.70892,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":174.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.89,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"fefde903c2faa46d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnest101e.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":10.5669,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":10.5842,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":94.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.89,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8667dbe5c8de2c29","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnest101e.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.89717,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.90057,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":526.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.89,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.06,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.54,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"be01383fb9b933f3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnest101e.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.11662,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.11951,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":320.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.89,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c7937261a8404137","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnest14d.gluon_in1k","title":"ResNeSt-14-D","source":"timm/resnest14d.gluon_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnest14d.gluon_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnest14d.gluon_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.541333,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.617249,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1755.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.93,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.25,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7ae8cbf549fe9c18","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnest14d.gluon_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.25981,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.26508,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":788.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.93,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.25,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"854379f4fd84a672","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnest14d.gluon_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.52694,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.53546,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":393.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.93,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.25,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2d52508722d18233","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnest14d.gluon_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.340793,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.342987,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2922.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":74.93,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.25,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.64,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.12,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"ac7a411a2232ca70","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnest14d.gluon_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.55135,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.553589,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1810.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.93,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.25,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"82617f3ff33a3f99","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnest26d.gluon_in1k","title":"ResNeSt-26-D","source":"timm/resnest26d.gluon_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnest26d.gluon_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnest26d.gluon_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.749517,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.881897,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1305.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.15,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"47e327c1f2fd40d4","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnest26d.gluon_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.8679,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.87604,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":532.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.15,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a763b494366441de","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnest26d.gluon_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.58081,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.58862,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":277.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.15,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2c9e48a7cd6931a1","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnest26d.gluon_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.506095,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.507874,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1970.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.15,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.71,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.19,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"9517dee415616637","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnest26d.gluon_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.890057,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.914062,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1122.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.15,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"07d9d85c50ad2ade","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnest50d.in1k","title":"ResNeSt-50-D","source":"timm/resnest50d.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnest50d.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnest50d.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.28406,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.42847,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":768.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.59,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.17,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4730a85383d03d5e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnest50d.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.94326,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.95209,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":337.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.59,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.17,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"fecfdc4a0ef794fc","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnest50d.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.51558,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.52557,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":180.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.59,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.17,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1c22347aedb8e3b4","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnest50d.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.84819,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.96706,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.850525,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.970276,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1177,"unit":"qps","digits":1,"baseline":1032.57,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.59,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.17,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.83,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.31,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"fa312842aa1a7a0c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnest50d.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.53122,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.54193,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":652.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.59,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.17,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"094da39a65730749","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnest50d_1s4x24d.in1k","title":"ResNeSt-50-D (1s4x24d)","source":"timm/resnest50d_1s4x24d.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnest50d_1s4x24d.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnest50d_1s4x24d.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.45132,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.54584,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":685.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.91,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.27,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4e36a2911d637140","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnest50d_1s4x24d.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.7082,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.71277,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":584.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.91,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.27,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.88,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.36,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"497337c70c7a2960","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnest50d_1s4x24d.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.99927,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.00916,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":499.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.91,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.27,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3c33822db9b6ac00","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnest50d_4s2x40d.in1k","title":"ResNeSt-50-D (4s2x40d)","source":"timm/resnest50d_4s2x40d.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnest50d_4s2x40d.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnest50d_4s2x40d.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.26134,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.36609,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":439.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.2,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.54,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"41ae92aaced4cf47","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnest50d_4s2x40d.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.81267,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.83038,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":206.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.2,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.54,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"548d1bddccdc6d20","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnest50d_4s2x40d.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.85479,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":9.87036,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":101,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.2,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.54,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6b2408f03fa02003","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnest50d_4s2x40d.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.3822,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.38654,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":419.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.2,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.54,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.94,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.42,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"8e8e18b8a35b1317","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnest50d_4s2x40d.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.93595,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.94501,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":340.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.2,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.54,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"efa7eccd261f8ef9","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet10t.c3_in1k","title":"ResNet-10-T","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet10t.c3_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet10t.c3_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":108.643,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":108.948,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":9060.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":65.98,"unit":"%","headline":true},{"description":"Top-5","value":86.5,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet10t.c3_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.107942,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.108307,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":9152.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":65.92,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":86.64,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.58,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":62.7,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"fbfe377fb07addc2","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/resnet101","title":"ResNet-101","source":"torchvision/resnet101","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fresnet101&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/resnet101","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.676570380164776,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":1478.0428308972882,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":79.482,"unit":"%","digits":2,"baseline":81.886,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":94.782,"unit":"%","digits":2,"baseline":95.78,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/resnet101_clip.openai","title":"ResNet-101 (CLIP)","source":"timm/resnet101_clip.openai","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet101_clip.openai&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet101_clip.openai","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.84042,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.07855,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.847107,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.08051,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1187.4,"unit":"qps","digits":1,"baseline":925.834,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.95,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.43,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"7c77cd16113f069b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet101_clip_gap.openai","title":"ResNet-101 (GAP, CLIP)","source":"timm/resnet101_clip_gap.openai","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet101_clip_gap.openai&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet101_clip_gap.openai","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.716909,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.964746,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.720337,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.967346,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1391.8,"unit":"qps","digits":1,"baseline":1035.02,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.84,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.32,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"7f1d15ced8263f6e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet101.a1_in1k","title":"ResNet-101 (timm)","source":"timm/resnet101.a1_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet101.a1_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet101.a1_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.50634,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.929304,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.508301,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.931549,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1968.8,"unit":"qps","digits":1,"baseline":1074.02,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.69,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.14,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.02,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.5,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"6e7f08448d3daa77","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet101c.gluon_in1k","title":"ResNet-101-C","source":"timm/resnet101c.gluon_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet101c.gluon_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet101c.gluon_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.719912,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.895422,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.800049,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.976013,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1366.1,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1094.31,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.27,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e782a452d18d370d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet101c.gluon_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.66998,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.67786,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":595.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.27,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2d40a4f37b5fd4ee","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet101c.gluon_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.92269,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.9425,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":340.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.27,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7fd0c173199de586","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet101c.gluon_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.634889,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.938324,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.636536,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.946838,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1571.4,"unit":"qps","digits":1,"baseline":1001.55,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.27,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.85,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.33,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"bd5b1e62736a0501","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet101c.gluon_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.12441,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.12701,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":888.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.27,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c7d5d53e595589e0","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet101d.gluon_in1k","title":"ResNet-101-D","source":"timm/resnet101d.gluon_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet101d.gluon_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet101d.gluon_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.740519,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.817932,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1325.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.52,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"9efa27de7ed88082","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet101d.gluon_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.88907,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.89728,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":526.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.52,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a84aa7edbe3fa719","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet101d.gluon_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.35457,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.36407,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":296.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.52,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"3b4270cdf128e3f6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet101d.gluon_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.723623,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.946417,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.725586,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.950378,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1378.7,"unit":"qps","digits":1,"baseline":1054.91,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.52,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.83,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.31,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"eb19d4523a036ba3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet101d.gluon_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.22971,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.23279,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":812.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.52,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1630629c02ea7519","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet14t.c3_in1k","title":"ResNet-14-T","source":"timm/resnet14t.c3_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet14t.c3_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet14t.c3_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.147581,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.179871,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":6656.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":70.67,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.27,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"139d327dbe34de28","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet14t.c3_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.551038,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.554688,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1800.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":70.67,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.27,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"738d083227daed67","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet14t.c3_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.92486,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.931091,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1073.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":70.67,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.27,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a88dc4ca25fcbd47","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet14t.c3_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.160907,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.162659,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":6106.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":70.67,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.27,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.63,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.11,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"948ddc7c2a348eb0","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet14t.c3_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.284454,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.286621,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3502.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":70.67,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.27,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"893aceb85467caf5","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/resnet152","title":"ResNet-152","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fresnet152&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/resnet152","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.824,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.827,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":547.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":81.82,"unit":"%","headline":true},{"description":"Top-5","value":95.75,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/resnet152","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.824,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.827,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":547.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":81.82,"unit":"%","headline":true},{"description":"Top-5","value":95.75,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet152.a1_in1k","title":"ResNet-152 (timm)","source":"timm/resnet152.a1_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet152.a1_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet152.a1_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.817566,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.32985,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.837341,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.33398,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1220.8,"unit":"qps","digits":1,"baseline":751.162,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.31,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.17,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.2,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.68,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"cd308436860e7553","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet152c.gluon_in1k","title":"ResNet-152-C","source":"timm/resnet152c.gluon_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet152c.gluon_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet152c.gluon_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.996465,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.10022,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":983.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.05,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.46,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2a0baef991cb5d5d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet152c.gluon_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.23992,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.24738,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":443.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.05,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.46,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f7eb3d46ac49c43d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet152c.gluon_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.85627,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.86542,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":257.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.05,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.46,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"47830d5b4a338799","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet152c.gluon_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.894804,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.34786,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.898376,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.3541,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1115.8,"unit":"qps","digits":1,"baseline":741.14,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.05,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.46,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.2,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.67,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"d19a48c7c9160831","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet152c.gluon_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.58472,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.58734,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":630.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.05,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.46,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b2cc79c5bde8cfd1","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/resnet18","title":"ResNet-18","source":"torchvision/resnet18","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fresnet18&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/resnet18","hardware":"thor","precision":"FP8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.148438,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.170166,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Throughput","value":6648.64,"unit":"qps","digits":1,"baseline":5803.21,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNette validation · 3,925 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":77.37579617834395,"unit":"%","digits":2,"baseline":77.93630573248407,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Top-5","value":94.80254777070064,"unit":"%","digits":2,"baseline":95.08280254777071,"baselineLabel":"Baseline TensorRT"}]},{"title":"Memory","metrics":[{"description":"Peak device memory","value":13,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":24,"baselineLabel":"Baseline TensorRT","improvementWord":"smaller","headline":true}]},{"title":"Size","metrics":[{"description":"Engine size","value":11.9,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":23.1,"baselineLabel":"Baseline TensorRT","improvementWord":"smaller","headline":true}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"torchvision/resnet18","hardware":"orin-nx-16gb","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"GPU compute, median","value":0.9547,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"p99 latency","value":0.9639,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1047.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNette val · 3,925 images · ImageNet labels · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":78.191,"unit":"%","digits":2,"headline":true},{"description":"Top-5","value":95.006,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Engine size","value":11.57,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Image","Classification"],"build":"fd700bd869c16cb6","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"torchvision/resnet18","hardware":"orin-nx-16gb","variant":"torchvision_int8_embedl-deploy_tensorrt-orin-nx-16gb","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"GPU compute, median","value":0.9547,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"p99 latency","value":0.9639,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1047.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNette val · 3,925 images · ImageNet labels · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":78.191,"unit":"%","digits":2,"headline":true},{"description":"Top-5","value":95.006,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Engine size","value":11.57,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Image","Classification"],"build":"fd700bd869c16cb6","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"torchvision/resnet18","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.112305,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.196777,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Throughput","value":8788.08,"unit":"qps","digits":1,"baseline":5036.71,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNette validation · 3,925 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":78.21656050955414,"unit":"%","digits":2,"baseline":78.0891719745223,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Top-5","value":95.13375796178345,"unit":"%","digits":2,"baseline":95.08280254777071,"baselineLabel":"Baseline TensorRT"}]},{"title":"Memory","metrics":[{"description":"Peak device memory","value":11,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":24,"baselineLabel":"Baseline TensorRT","improvementWord":"smaller","headline":true}]},{"title":"Size","metrics":[{"description":"Engine size","value":13.8,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":23.1,"baselineLabel":"Baseline TensorRT","improvementWord":"smaller","headline":true}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"torchvision/resnet18","hardware":"rtx-2080-ti","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.2602,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"p99 latency","value":0.2636,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3840.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNette validation · 3,925 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":78.242,"unit":"%","digits":2,"headline":true},{"description":"Top-5","value":94.981,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Engine size","value":11.57,"unit":"MiB","higherIsBetter":false,"digits":2,"headline":true}]}],"labels":["Image","Classification"],"build":"2026101321097e0a","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"torchvision/resnet18","hardware":"rtx-2080-ti","variant":"torchvision_int8_embedl-deploy_tensorrt-rtx-2080-ti","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.2602,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"p99 latency","value":0.2636,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3840.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNette validation · 3,925 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":78.242,"unit":"%","digits":2,"headline":true},{"description":"Top-5","value":94.981,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Engine size","value":11.57,"unit":"MiB","higherIsBetter":false,"digits":2,"headline":true}]}],"labels":["Image","Classification"],"build":"2026101321097e0a","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/resnet18.a1_in1k","title":"ResNet-18 (timm)","source":"timm/resnet18.a1_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet18.a1_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet18.a1_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.104374,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.183916,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.105286,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.185059,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":9442.6,"unit":"qps","digits":1,"baseline":5398.14,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":70.78,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.47,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.65,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.13,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"4ff49a92f1af6d5e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet18d.ra2_in1k","title":"ResNet-18-D","source":"timm/resnet18d.ra2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet18d.ra2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet18d.ra2_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.174423,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.225555,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":5542,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":71.98,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.51,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"21f7842ad36ea83f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet18d.ra2_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.492446,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.496155,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2014.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":71.98,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.51,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"06352a2f993b7355","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet18d.ra2_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.798552,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.80304,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1243.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":71.98,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.51,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"61cf88be4fb0c610","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet18d.ra2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.163553,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.195703,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.164368,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.196411,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":6055.5,"unit":"qps","digits":1,"baseline":5074.65,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":71.98,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.51,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.65,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.13,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"be13b9ffc8539a81","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet18d.ra2_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.368143,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.370697,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2708.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":71.98,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.51,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"bdd9b25ab564b4ec","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet200d.ra2_in1k","title":"ResNet-200-D","source":"timm/resnet200d.ra2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet200d.ra2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet200d.ra2_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.28531,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.40948,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":770.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.31,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"83a8ec05830f3698","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet200d.ra2_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.98801,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.99695,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":332.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.31,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"eecb10dc976dd2ee","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet200d.ra2_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.15759,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.16833,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":192.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.31,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ca6cb21ad60f3ee8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet200d.ra2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.19775,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.78185,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.20251,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.78522,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":833.9,"unit":"qps","digits":1,"baseline":560.773,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.31,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.25,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.73,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"0090ff921f64228c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet200d.ra2_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.07474,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.07874,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":481.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.31,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"493a9e6bf6866766","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet26.bt_in1k","title":"ResNet-26","source":"timm/resnet26.bt_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet26.bt_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet26.bt_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.789,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.796082,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1258.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.23,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2934256c2b9eb495","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet26.bt_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.62627,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.63318,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":611.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.23,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"58cd06d88db1dab5","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet26.bt_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.202642,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.228292,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.203552,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.229187,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":4902.3,"unit":"qps","digits":1,"baseline":4353.97,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":74.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.23,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.7,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.18,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"12d1081093780198","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet26.bt_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.397998,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.400452,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2506.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.23,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"954e991ad1c38625","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet26d.bt_in1k","title":"ResNet-26-D","source":"timm/resnet26d.bt_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet26d.bt_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet26d.bt_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.247909,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.296631,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3914.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.19,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.96,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"56e2a85ffed3dfde","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet26d.bt_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.00543,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.01086,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":988.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.19,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.96,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"edc58ee0b9648d54","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet26d.bt_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.89662,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.90332,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":524.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.19,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.96,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"98bea5f67eab7250","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet26d.bt_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.291161,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.293274,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3418,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.19,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.96,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.7,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.18,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e7da444ccbc69e6d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet26d.bt_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.510036,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.512207,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1956.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.19,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.96,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"ee0393e5242c18bc","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet26t.ra2_in1k","title":"ResNet-26-T","source":"timm/resnet26t.ra2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet26t.ra2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet26t.ra2_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.254804,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.297119,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3787.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.62,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2a96191a67aafdba","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet26t.ra2_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.910209,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.91687,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1091.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.62,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"cf595110f0e71a81","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet26t.ra2_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.79866,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.80475,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":552.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.62,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"98486a7ea3ea7c99","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet26t.ra2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.279661,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.28064,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3556.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.62,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.7,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.18,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"c2e325f868a6b106","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet26t.ra2_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.500578,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.503479,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1993.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.62,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"25b52db29b19ddb7","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet32ts.ra2_in1k","title":"ResNet-32-TS","source":"timm/resnet32ts.ra2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet32ts.ra2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet32ts.ra2_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.471281,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.523987,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2025.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.18,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f5cd721edcdd86f3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet32ts.ra2_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.20041,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.20746,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":827.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.18,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b023e3111bd0c26b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet32ts.ra2_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.17996,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.19354,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":456.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.18,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"085433c9423d1441","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet32ts.ra2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.1929,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.19696,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":837.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.18,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.78,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.26,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"b8c064a85a1b932c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet32ts.ra2_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.40541,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.41492,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":711,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.18,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1bd03035dc37e72f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet33ts.ra2_in1k","title":"ResNet-33-TS","source":"timm/resnet33ts.ra2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet33ts.ra2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet33ts.ra2_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.490393,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.544678,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1942.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.72,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.34,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2c139e836310950f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet33ts.ra2_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.34241,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.35175,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":740.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.72,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.34,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"1c83665bfa6e1000","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet33ts.ra2_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.58585,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.59448,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":384.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.72,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.34,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b642720b7672eb12","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet33ts.ra2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.24195,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.24652,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":804.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.72,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.34,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.81,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.28,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"878a437cb389695c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet33ts.ra2_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.48265,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.49249,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":674,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.72,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.34,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"738368adbb87c7b6","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/resnet34","title":"ResNet-34","source":"torchvision/resnet34","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fresnet34&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/resnet34","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.3709942661225796,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":2695.4594486093533,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":72.978,"unit":"%","digits":2,"baseline":73.314,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":91.29,"unit":"%","digits":2,"baseline":91.42,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/resnet34.a1_in1k","title":"ResNet-34 (timm)","source":"timm/resnet34.a1_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet34.a1_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet34.a1_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.192526,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.359285,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.200043,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.360596,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":5147.6,"unit":"qps","digits":1,"baseline":2770.04,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":75.89,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.5,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.77,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.24,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"1bd7dab1cd15a75a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet34d.ra2_in1k","title":"ResNet-34-D","source":"timm/resnet34d.ra2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet34d.ra2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet34d.ra2_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.266679,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.417357,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.324036,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.463013,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":3644.9,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":2266.93,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.21,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b15b7b49eb11950e","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet34d.ra2_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.734097,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.73877,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1352.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.21,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"abe529286f25d2ac","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet34d.ra2_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.16569,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.17188,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":852.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.21,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"edfcafba9112ccd5","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet34d.ra2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.250981,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.370188,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.251831,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.371216,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":3960.3,"unit":"qps","digits":1,"baseline":2691.02,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.21,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.77,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.24,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"01d2b0a3a23cd24b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet34d.ra2_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.608197,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.611877,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1641.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.21,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"61b2b2dda0283cf9","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/resnet50","title":"ResNet-50","source":"torchvision/resnet50","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fresnet50&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/resnet50","hardware":"thor","precision":"FP8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.325195,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.596924,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Throughput","value":3054.87,"unit":"qps","digits":1,"baseline":1656.01,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNette validation · 3,925 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":85.68152866242038,"unit":"%","digits":2,"baseline":86.31847133757961,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Top-5","value":98.44585987261146,"unit":"%","digits":2,"baseline":98.59872611464968,"baselineLabel":"Baseline TensorRT"}]},{"title":"Memory","metrics":[{"description":"Peak device memory","value":26,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":53,"baselineLabel":"Baseline TensorRT","improvementWord":"smaller","headline":true}]},{"title":"Size","metrics":[{"description":"Engine size","value":25.5,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":49.9,"baselineLabel":"Baseline TensorRT","improvementWord":"smaller","headline":true}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"torchvision/resnet50","hardware":"orin-nano-super","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.72314,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":579.65,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNette validation · 3,925 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":84.71337579617835,"unit":"%","digits":2,"baseline":86.36942675159236,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":98.31847133757961,"unit":"%","digits":2,"baseline":98.59872611464968,"baselineLabel":"FP32 model"}]},{"title":"Size","metrics":[{"description":"Engine size","value":25,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"torchvision/resnet50","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.46242107986472547,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":2162.5311724381927,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":79.428,"unit":"%","digits":2,"baseline":80.858,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":94.75200000000001,"unit":"%","digits":2,"baseline":95.434,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/resnet50_clip.cc12m","title":"ResNet-50 (CLIP, CC12M)","source":"timm/resnet50_clip.cc12m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet50_clip.cc12m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet50_clip.cc12m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.508467,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.675446,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.510742,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.678223,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1960,"unit":"qps","digits":1,"baseline":1477.51,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.95,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.43,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"0392b08fc261ce7d","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet50_clip.openai","title":"ResNet-50 (CLIP, OpenAI)","source":"timm/resnet50_clip.openai","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet50_clip.openai&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet50_clip.openai","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.474297,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.674062,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.476227,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.676697,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":2101.7,"unit":"qps","digits":1,"baseline":1479.79,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.81,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.29,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"22064c4d0624b315","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet50_clip_gap.cc12m","title":"ResNet-50 (GAP, CLIP, CC12M)","source":"timm/resnet50_clip_gap.cc12m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet50_clip_gap.cc12m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet50_clip_gap.cc12m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.407741,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.491697,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.409607,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.493835,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":2443.1,"unit":"qps","digits":1,"baseline":2027.97,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.79,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.26,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"325e56a8d7d43b8b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet50_clip_gap.openai","title":"ResNet-50 (GAP, CLIP, OpenAI)","source":"timm/resnet50_clip_gap.openai","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet50_clip_gap.openai&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet50_clip_gap.openai","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.412458,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.499886,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.4151,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.501404,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":2414.9,"unit":"qps","digits":1,"baseline":1994.28,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.7,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.18,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"3adcfbd25091d0e6","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet50.a1_in1k","title":"ResNet-50 (timm)","source":"timm/resnet50.a1_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet50.a1_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet50.a1_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.245632,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.470524,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.246582,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.47229,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":4045.6,"unit":"qps","digits":1,"baseline":2119.13,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.52,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.71,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.81,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.29,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"88ac2411d3a15ec6","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet50c.gluon_in1k","title":"ResNet-50-C","source":"timm/resnet50c.gluon_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet50c.gluon_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet50c.gluon_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.352131,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.409485,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2765.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.87,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"175ee9fd51f620d6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet50c.gluon_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.08871,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.09937,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":912.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.87,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"611318180f4cfc96","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet50c.gluon_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.94687,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.95459,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":510.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.87,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f38e66181df624e2","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet50c.gluon_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.317211,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.489661,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.317993,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.490967,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":3138.9,"unit":"qps","digits":1,"baseline":2036.26,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.87,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.7,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.18,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"4929868755680e86","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet50c.gluon_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.642399,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.645142,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1554.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.87,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"70b30b98279ad45b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet50d.a1_in1k","title":"ResNet-50-D","source":"timm/resnet50d.a1_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet50d.a1_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet50d.a1_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.380062,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.548101,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.44339,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.610352,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":2574.7,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1751.9,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.57,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"265113fc78fbf6a1","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet50d.a1_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.01265,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.01837,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":981.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.57,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"53667ef25757f650","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet50d.a1_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.78911,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.79633,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":555.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.57,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a7350446423beba0","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet50d.a1_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.323591,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.484945,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.324463,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.486328,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":3075.7,"unit":"qps","digits":1,"baseline":2056.2,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.57,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.81,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.29,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"51a2c4acb77e6bd3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet50d.a1_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.645913,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.649048,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1545.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":80.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.57,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"394f9ae9ba6b9e2b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet50s.gluon_in1k","title":"ResNet-50-S","source":"timm/resnet50s.gluon_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet50s.gluon_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet50s.gluon_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.34625,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.539524,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.347473,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.541565,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":2875.1,"unit":"qps","digits":1,"baseline":1848.64,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.05,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.45,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.71,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.19,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"4b5c21a79411fc93","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet50x16_clip.openai","title":"ResNet-50x16 (CLIP)","source":"timm/resnet50x16_clip.openai","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet50x16_clip.openai&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet50x16_clip.openai","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.56442,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.88133,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":2.64972,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.98718,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":389.7,"unit":"qps","digits":1,"baseline":257.539,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.78,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.26,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"c880ca6bd16a0a44","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet50x16_clip_gap.openai","title":"ResNet-50x16 (GAP, CLIP)","source":"timm/resnet50x16_clip_gap.openai","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet50x16_clip_gap.openai&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet50x16_clip_gap.openai","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.35979,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.56446,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":2.48053,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.75195,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":423.5,"unit":"qps","digits":1,"baseline":280.42,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.55,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.03,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"31c087a037e05da4","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet50x4_clip.openai","title":"ResNet-50x4 (CLIP)","source":"timm/resnet50x4_clip.openai","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet50x4_clip.openai&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet50x4_clip.openai","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.20853,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.75337,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.2215,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.78442,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":826.4,"unit":"qps","digits":1,"baseline":569.779,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.18,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.66,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"252040768533c917","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet50x4_clip_gap.openai","title":"ResNet-50x4 (GAP, CLIP)","source":"timm/resnet50x4_clip_gap.openai","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet50x4_clip_gap.openai&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet50x4_clip_gap.openai","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.14016,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.52255,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.14331,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.52631,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":875.8,"unit":"qps","digits":1,"baseline":656.192,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.02,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.5,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"c7fd9cf4c7bc17d1","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet50x64_clip.openai","title":"ResNet-50x64 (CLIP)","source":"timm/resnet50x64_clip.openai","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet50x64_clip.openai&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet50x64_clip.openai","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.9818,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":10.9989,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":7.13599,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":11.4429,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":143.2,"unit":"qps","digits":1,"baseline":90.9052,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.68,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.16,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"921e2e7c16551dc2","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet50x64_clip_gap.openai","title":"ResNet-50x64 (GAP, CLIP)","source":"timm/resnet50x64_clip_gap.openai","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet50x64_clip_gap.openai&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet50x64_clip_gap.openai","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.6036,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":10.3121,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":6.78687,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":10.6124,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":151.4,"unit":"qps","digits":1,"baseline":96.9596,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.26,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.74,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"98b5e649f631730e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet51q.ra2_in1k","title":"ResNet-51-Q","source":"timm/resnet51q.ra2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet51q.ra2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet51q.ra2_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.59492,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.61029,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":276.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.58,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.85,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"202610202012b74f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/resnet51q.ra2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.41329,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.41602,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":706.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.58,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.85,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.02,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.5,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"80147de8d472d708","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnet61q.ra2_in1k","title":"ResNet-61-Q","source":"timm/resnet61q.ra2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnet61q.ra2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnet61q.ra2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.51834,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.52643,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":658,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.87,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.85,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.03,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.51,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"0cd9dfd3d312ade0","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnetaa50.a1h_in1k","title":"ResNet-AA-50","source":"timm/resnetaa50.a1h_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnetaa50.a1h_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnetaa50.a1h_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.324191,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.497105,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.3255,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.501404,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":3071.6,"unit":"qps","digits":1,"baseline":2005.83,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.92,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.87,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.81,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.29,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"48470a234bdc0cc8","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnetblur50.bt_in1k","title":"ResNet-Blur-50","source":"timm/resnetblur50.bt_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnetblur50.bt_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnetblur50.bt_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.406726,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.560334,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.408997,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.562683,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":2450.2,"unit":"qps","digits":1,"baseline":1780.34,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.8,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.85,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.81,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.29,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"358519dbaa11d0b2","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnetrs101.tf_in1k","title":"ResNet-RS-101","source":"timm/resnetrs101.tf_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnetrs101.tf_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnetrs101.tf_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.4763,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.47821,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":676.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.81,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.81,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.23,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.71,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"69f5b8698e99d152","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnetrs200.tf_in1k","title":"ResNet-RS-200","source":"timm/resnetrs200.tf_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnetrs200.tf_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnetrs200.tf_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.05073,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.09241,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":327.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.14,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.56,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.04,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"d925195c02ada43d","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnetrs270.tf_in1k","title":"ResNet-RS-270","source":"timm/resnetrs270.tf_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnetrs270.tf_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnetrs270.tf_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.1508,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.21704,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":240.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.18,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.28,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.96,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.44,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"73a1bb0f7b4d4af9","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnetrs350.tf_in1k","title":"ResNet-RS-350","source":"timm/resnetrs350.tf_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnetrs350.tf_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnetrs350.tf_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.98177,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.06396,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":167.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.41,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.04,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.34,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.82,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"f6d5239b509c332b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnetrs420.tf_in1k","title":"ResNet-RS-420","source":"timm/resnetrs420.tf_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnetrs420.tf_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnetrs420.tf_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.5535,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":7.64258,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":132.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.65,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.71,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.65,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.13,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"aa89b09902d332a6","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnetrs50.tf_in1k","title":"ResNet-RS-50","source":"timm/resnetrs50.tf_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnetrs50.tf_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnetrs50.tf_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.665147,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.666443,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1500,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.08,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.41,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.4,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"90ec703651a8a680","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnetv2_18d.ra4_e3600_r224_in1k","title":"ResNetV2 (18d)","source":"timm/resnetv2_18d.ra4_e3600_r224_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnetv2_18d.ra4_e3600_r224_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnetv2_18d.ra4_e3600_r224_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.230401,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.23114,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":4312,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":73.69,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.49,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.65,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.13,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"4b417b1fabe34c4f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnetv2_34d.ra4_e3600_r224_in1k","title":"ResNetV2 (34d)","source":"timm/resnetv2_34d.ra4_e3600_r224_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnetv2_34d.ra4_e3600_r224_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnetv2_34d.ra4_e3600_r224_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.422705,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.423889,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2358.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.96,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.72,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.77,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.24,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"40ceaf7a193a5a0a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnetv2_50d_gn.ah_in1k","title":"ResNetV2 (50d, gn)","source":"timm/resnetv2_50d_gn.ah_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnetv2_50d_gn.ah_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnetv2_50d_gn.ah_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.912274,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.06978,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.914795,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.07208,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1094.2,"unit":"qps","digits":1,"baseline":933.552,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.46,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.19,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.81,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.29,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"bda23d16ca8f406c","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnetv2_101.a1h_in1k","title":"ResNetV2-101","source":"timm/resnetv2_101.a1h_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnetv2_101.a1h_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnetv2_101.a1h_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.16173,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.16461,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":859.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.39,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.6,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.02,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.5,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"c23a7e91acefb7f6","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnetv2_101x1_bit.goog_in21k_ft_in1k","title":"ResNetV2-101x1 (BiT)","source":"timm/resnetv2_101x1_bit.goog_in21k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnetv2_101x1_bit.goog_in21k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnetv2_101x1_bit.goog_in21k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.6136,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.18786,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":3.73462,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.30554,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":276.6,"unit":"qps","digits":1,"baseline":238.705,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.71,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.35,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.03,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.51,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"7b239f3b8560a682","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnetv2_101x3_bit.goog_in21k_ft_in1k","title":"ResNetV2-101x3 (BiT)","source":"timm/resnetv2_101x3_bit.goog_in21k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnetv2_101x3_bit.goog_in21k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnetv2_101x3_bit.goog_in21k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":17.4511,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":23.4614,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":18.0887,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":24.0546,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":57.3,"unit":"qps","digits":1,"baseline":42.6205,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.43,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.39,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4.9,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":4.38,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"0212615268012367","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnetv2_152x2_bit.goog_in21k_ft_in1k","title":"ResNetV2-152x2 (BiT)","source":"timm/resnetv2_152x2_bit.goog_in21k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnetv2_152x2_bit.goog_in21k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnetv2_152x2_bit.goog_in21k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":12.3209,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":16.2173,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":12.6101,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":16.8055,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":81.2,"unit":"qps","digits":1,"baseline":61.6561,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.46,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.4,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.19,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.66,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"89f1bdc8d62e205f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnetv2_152x4_bit.goog_in21k_ft_in1k","title":"ResNetV2-152x4 (BiT)","source":"timm/resnetv2_152x4_bit.goog_in21k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnetv2_152x4_bit.goog_in21k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnetv2_152x4_bit.goog_in21k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":49.0861,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":78.264,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":50.5229,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":80.0563,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":20.4,"unit":"qps","digits":1,"baseline":12.777,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.55,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.27,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":11.03,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":10.51,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"ff2e3fe55331fd29","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnetv2_18.ra4_e3600_r224_in1k","title":"ResNetV2-18","source":"timm/resnetv2_18.ra4_e3600_r224_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnetv2_18.ra4_e3600_r224_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnetv2_18.ra4_e3600_r224_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.212946,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.21405,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":4660.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":72.8,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.11,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.65,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.13,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"30b2436f596d8778","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnetv2_34.ra4_e3600_r224_in1k","title":"ResNetV2-34","source":"timm/resnetv2_34.ra4_e3600_r224_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnetv2_34.ra4_e3600_r224_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnetv2_34.ra4_e3600_r224_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.394337,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.395447,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2526.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.26,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.77,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.24,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"fe59afc70f221c0e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnetv2_50.a1h_in1k","title":"ResNetV2-50","source":"timm/resnetv2_50.a1h_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnetv2_50.a1h_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnetv2_50.a1h_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.559998,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.562378,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1780.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.65,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.71,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.81,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.29,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"8fce2b7d6135a81e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnetv2_50x1_bit.goog_distilled_in1k","title":"ResNetV2-50x1 (BiT)","source":"timm/resnetv2_50x1_bit.goog_distilled_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnetv2_50x1_bit.goog_distilled_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnetv2_50x1_bit.goog_distilled_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.02369,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.64371,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.02673,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.65106,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":975.3,"unit":"qps","digits":1,"baseline":607.874,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.02,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.2,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.82,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.29,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"3e9f9ca271eba17d","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnetv2_50x3_bit.goog_in21k_ft_in1k","title":"ResNetV2-50x3 (BiT)","source":"timm/resnetv2_50x3_bit.goog_in21k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnetv2_50x3_bit.goog_in21k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnetv2_50x3_bit.goog_in21k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":11.6069,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":13.9487,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":11.916,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":14.4934,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":86.1,"unit":"qps","digits":1,"baseline":71.6837,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.98,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.49,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"6c4500ae5d01b8ca","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnext101_32x32d.fb_wsl_ig1b_ft_in1k","title":"ResNeXt-101 32x32d","source":"timm/resnext101_32x32d.fb_wsl_ig1b_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnext101_32x32d.fb_wsl_ig1b_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnext101_32x32d.fb_wsl_ig1b_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.99284,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":6.45659,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":3.05249,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":6.62244,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":333.9,"unit":"qps","digits":1,"baseline":154.844,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.09,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.07,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":5.76,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":5.23,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"db7fe77326d17d40","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnext101_32x4d.fb_ssl_yfcc100m_ft_in1k","title":"ResNeXt-101 32x4d","source":"timm/resnext101_32x4d.fb_ssl_yfcc100m_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnext101_32x4d.fb_ssl_yfcc100m_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnext101_32x4d.fb_ssl_yfcc100m_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.609304,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.872458,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.611694,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.875061,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1637.7,"unit":"qps","digits":1,"baseline":1143.89,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.85,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.22,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.02,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.5,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"017228ccf8eae87a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/resnext101_32x8d","title":"ResNeXt-101 32x8d","source":"torchvision/resnext101_32x8d","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fresnext101_32x8d&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/resnext101_32x8d","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.0222685616463423,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":978.2165250093584,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":81.76400000000001,"unit":"%","digits":2,"baseline":82.834,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":95.878,"unit":"%","digits":2,"baseline":96.228,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"torchvision/resnext101_64x4d","title":"ResNeXt-101 64x4d","source":"torchvision/resnext101_64x4d","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fresnext101_64x4d&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/resnext101_64x4d","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.0476220399141312,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":954.5427281025565,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":82.338,"unit":"%","digits":2,"baseline":83.246,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":95.968,"unit":"%","digits":2,"baseline":96.454,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/resnext101_64x4d.c1_in1k","title":"ResNeXt-101 64x4d (timm)","source":"timm/resnext101_64x4d.c1_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnext101_64x4d.c1_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnext101_64x4d.c1_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.894182,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.4321,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.898132,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.43787,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1116.7,"unit":"qps","digits":1,"baseline":697.584,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.92,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.78,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.46,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.93,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"c8b44e188ae9ffc0","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnext26ts.ra2_in1k","title":"ResNeXt-26-TS","source":"timm/resnext26ts.ra2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnext26ts.ra2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnext26ts.ra2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.557404,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.559875,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1789,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.58,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.02,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.66,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.14,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"f525c97b8f479429","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/resnext50_32x4d","title":"ResNeXt-50 32x4d","source":"torchvision/resnext50_32x4d","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fresnext50_32x4d&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/resnext50_32x4d","hardware":"orin-nano-super","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.59985,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":624.296,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNette validation · 3,925 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":88.84076433121018,"unit":"%","digits":2,"baseline":89.45222929936307,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":99.3375796178344,"unit":"%","digits":2,"baseline":99.3375796178344,"baselineLabel":"FP32 model"}]},{"title":"Size","metrics":[{"description":"Engine size","value":25,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"torchvision/resnext50_32x4d","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.43018659576773643,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":2324.5726618128137,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":80.502,"unit":"%","digits":2,"baseline":81.198,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":95.112,"unit":"%","digits":2,"baseline":95.34,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/resnext50_32x4d.a1_in1k","title":"ResNeXt-50 32x4d (timm)","source":"timm/resnext50_32x4d.a1_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnext50_32x4d.a1_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnext50_32x4d.a1_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.241622,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.438527,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.242249,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.440186,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":4109.1,"unit":"qps","digits":1,"baseline":2272.34,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.95,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.87,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.8,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.28,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"8e1b7455eeb58fef","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/resnext50d_32x4d.bt_in1k","title":"ResNeXt-50-D 32x4d","source":"timm/resnext50d_32x4d.bt_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fresnext50d_32x4d.bt_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/resnext50d_32x4d.bt_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.351226,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.466295,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.352417,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.468262,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":2834.8,"unit":"qps","digits":1,"baseline":2138.01,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.6,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.71,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.8,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.28,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"c6a5bb26c5895392","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/rexnet_100.nav_in1k","title":"ReXNet 1.0×","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Frexnet_100.nav_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/rexnet_100.nav_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":563.538,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":564.758,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":1769.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.21,"unit":"%","headline":true},{"description":"Top-5","value":0.72,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/rexnet_130.nav_in1k","title":"ReXNet 1.3×","source":"timm/rexnet_130.nav_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Frexnet_130.nav_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/rexnet_130.nav_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.742379,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.744751,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1344.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":5.22,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":14.33,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.61,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.09,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"7b31687e74646163","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"embedl/rf-detr-base","upstream":"roboflow/rf-detr","title":"RF-DETR Base","upstreamUrl":"https://github.com/roboflow/rf-detr","source":"embedl/rf-detr-base","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=embedl%2Frf-detr-base-quantized-tensorrt&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"embedl/rf-detr-base","hardware":"thor","precision":"INT8","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":2.0822079181671143,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.2954399585723877,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Accuracy","subtitle":"COCO val2017","metrics":[{"description":"mAP","value":52.4,"unit":"%","digits":2,"baseline":53.29,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"mAP@50","value":71.43,"unit":"%","digits":2,"baseline":72.2,"baselineLabel":"Baseline TensorRT"},{"description":"mAP@75","value":56.9,"unit":"%","digits":2,"baseline":57.57,"baselineLabel":"Baseline TensorRT"}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":546.8,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":544.5,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":37.85,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":54.8,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image","Detection"],"build":"003ecfa7e6894852","updatedAt":"2026-11-04T20:59:53.000Z"},{"model":"embedl/rf-detr-base","hardware":"thor","variant":"huggingface_int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":2.0822079181671143,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.2954399585723877,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Accuracy","subtitle":"COCO val2017","metrics":[{"description":"mAP","value":52.4,"unit":"%","digits":2,"baseline":53.29,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"mAP@50","value":71.43,"unit":"%","digits":2,"baseline":72.2,"baselineLabel":"Baseline TensorRT"},{"description":"mAP@75","value":56.9,"unit":"%","digits":2,"baseline":57.57,"baselineLabel":"Baseline TensorRT"}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":546.8,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":544.5,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":37.85,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":54.8,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image","Detection"],"build":"003ecfa7e6894852","updatedAt":"2026-11-04T20:59:53.000Z"},{"model":"embedl/rf-detr-base","hardware":"thor","variant":"rfdetr_fp16_tensorrt","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":2.2954399585723877,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true}]},{"title":"Accuracy","subtitle":"COCO val2017","metrics":[{"description":"mAP","value":53.29,"unit":"%","digits":2,"headline":true},{"description":"mAP@50","value":72.2,"unit":"%","digits":2},{"description":"mAP@75","value":57.57,"unit":"%","digits":2}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":544.5,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":54.8,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image","Detection"],"build":"1baef8b9a091d314","updatedAt":"2026-11-04T19:23:15.000Z"},{"model":"embedl/rf-detr-base","hardware":"orin","precision":"INT8","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":5.950911998748779,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":6.909168004989624,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Accuracy","subtitle":"COCO val2017","metrics":[{"description":"mAP","value":52.44,"unit":"%","digits":2,"baseline":53.3,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"mAP@50","value":71.61,"unit":"%","digits":2,"baseline":72.15,"baselineLabel":"Baseline TensorRT"},{"description":"mAP@75","value":56.84,"unit":"%","digits":2,"baseline":57.64,"baselineLabel":"Baseline TensorRT"}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":631,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":621,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":37.98,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":54.7,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image","Detection"],"build":"695c5a2286d94889","updatedAt":"2026-11-04T20:59:53.000Z"},{"model":"embedl/rf-detr-base","hardware":"orin","variant":"huggingface_int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":5.950911998748779,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":6.909168004989624,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Accuracy","subtitle":"COCO val2017","metrics":[{"description":"mAP","value":52.44,"unit":"%","digits":2,"baseline":53.3,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"mAP@50","value":71.61,"unit":"%","digits":2,"baseline":72.15,"baselineLabel":"Baseline TensorRT"},{"description":"mAP@75","value":56.84,"unit":"%","digits":2,"baseline":57.64,"baselineLabel":"Baseline TensorRT"}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":631,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":621,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":37.98,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":54.7,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image","Detection"],"build":"695c5a2286d94889","updatedAt":"2026-11-04T20:59:53.000Z"},{"model":"embedl/rf-detr-base","hardware":"orin","variant":"rfdetr_fp16_tensorrt","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":6.909168004989624,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true}]},{"title":"Accuracy","subtitle":"COCO val2017","metrics":[{"description":"mAP","value":53.3,"unit":"%","digits":2,"headline":true},{"description":"mAP@50","value":72.15,"unit":"%","digits":2},{"description":"mAP@75","value":57.64,"unit":"%","digits":2}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":621,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":54.7,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image","Detection"],"build":"e1632122a1d6e152","updatedAt":"2026-11-04T10:43:05.000Z"},{"model":"embedl/rf-detr-base","hardware":"orin-nx-16gb","precision":"INT8","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":11.564080238342285,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":13.26540756225586,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Accuracy","subtitle":"COCO val2017","metrics":[{"description":"mAP","value":52.37,"unit":"%","digits":2,"baseline":53.3,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"mAP@50","value":71.53,"unit":"%","digits":2,"baseline":72.15,"baselineLabel":"Baseline TensorRT"},{"description":"mAP@75","value":56.72,"unit":"%","digits":2,"baseline":57.56,"baselineLabel":"Baseline TensorRT"}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":663.2,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":640.5,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":37.92,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":56.39,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image","Detection"],"build":"3fbb1c55c87c847b","updatedAt":"2026-11-04T20:59:53.000Z"},{"model":"embedl/rf-detr-base","hardware":"orin-nx-16gb","variant":"huggingface_int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":11.564080238342285,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":13.26540756225586,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Accuracy","subtitle":"COCO val2017","metrics":[{"description":"mAP","value":52.37,"unit":"%","digits":2,"baseline":53.3,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"mAP@50","value":71.53,"unit":"%","digits":2,"baseline":72.15,"baselineLabel":"Baseline TensorRT"},{"description":"mAP@75","value":56.72,"unit":"%","digits":2,"baseline":57.56,"baselineLabel":"Baseline TensorRT"}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":663.2,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":640.5,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":37.92,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":56.39,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image","Detection"],"build":"3fbb1c55c87c847b","updatedAt":"2026-11-04T20:59:53.000Z"},{"model":"embedl/rf-detr-base","hardware":"orin-nx-16gb","variant":"rfdetr_fp16_tensorrt","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":13.26540756225586,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true}]},{"title":"Accuracy","subtitle":"COCO val2017","metrics":[{"description":"mAP","value":53.3,"unit":"%","digits":2,"headline":true},{"description":"mAP@50","value":72.15,"unit":"%","digits":2},{"description":"mAP@75","value":57.56,"unit":"%","digits":2}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":640.5,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":56.39,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image","Detection"],"build":"ad7d8b3d86e388cd","updatedAt":"2026-11-04T10:43:05.000Z"},{"model":"embedl/rf-detr-base","hardware":"orin-nano-super","precision":"INT8","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":13.038175582885742,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":16.3371524810791,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Accuracy","subtitle":"COCO val2017","metrics":[{"description":"mAP","value":52.36,"unit":"%","digits":2,"baseline":53.28,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"mAP@50","value":71.52,"unit":"%","digits":2,"baseline":72.15,"baselineLabel":"Baseline TensorRT"},{"description":"mAP@75","value":56.72,"unit":"%","digits":2,"baseline":57.59,"baselineLabel":"Baseline TensorRT"}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":642.1,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":737.1,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":38.23,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":56.38,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image","Detection"],"build":"01959115d25d103c","updatedAt":"2026-11-04T20:59:53.000Z"},{"model":"embedl/rf-detr-base","hardware":"orin-nano-super","variant":"huggingface_int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":13.038175582885742,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":16.3371524810791,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Accuracy","subtitle":"COCO val2017","metrics":[{"description":"mAP","value":52.36,"unit":"%","digits":2,"baseline":53.28,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"mAP@50","value":71.52,"unit":"%","digits":2,"baseline":72.15,"baselineLabel":"Baseline TensorRT"},{"description":"mAP@75","value":56.72,"unit":"%","digits":2,"baseline":57.59,"baselineLabel":"Baseline TensorRT"}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":642.1,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":737.1,"baselineLabel":"Baseline TensorRT","headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":38.23,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":56.38,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image","Detection"],"build":"01959115d25d103c","updatedAt":"2026-11-04T20:59:53.000Z"},{"model":"embedl/rf-detr-base","hardware":"orin-nano-super","variant":"rfdetr_fp16_tensorrt","precision":"FP16","metrics":[{"title":"Latency","metrics":[{"description":"Median GPU time","value":16.3371524810791,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true}]},{"title":"Accuracy","subtitle":"COCO val2017","metrics":[{"description":"mAP","value":53.28,"unit":"%","digits":2,"headline":true},{"description":"mAP@50","value":72.15,"unit":"%","digits":2},{"description":"mAP@75","value":57.59,"unit":"%","digits":2}]},{"title":"Memory","metrics":[{"description":"Peak memory","value":737.1,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]},{"title":"Size","metrics":[{"description":"Model size","value":56.38,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true},{"description":"Container size","value":703.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image","Detection"],"build":"442bec2dede767bf","updatedAt":"2026-11-04T10:43:05.000Z"}]},{"name":"timm/sam2_hiera_base_plus.fb_r896","title":"SAM 2 Hiera Base Plus","source":"timm/sam2_hiera_base_plus.fb_r896","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fsam2_hiera_base_plus.fb_r896&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/sam2_hiera_base_plus.fb_r896","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":40.9535,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":41.5854,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":24.4,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.31,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.79,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"5ec05e8981d59ce7","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/sam2_hiera_large.fb_r1024","title":"SAM 2 Hiera Large","source":"timm/sam2_hiera_large.fb_r1024","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fsam2_hiera_large.fb_r1024&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/sam2_hiera_large.fb_r1024","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":104.42,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":107.572,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":9.6,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.94,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.42,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"e8160fd7ed425aee","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/sam2_hiera_small.fb_r896","title":"SAM 2 Hiera Small","source":"timm/sam2_hiera_small.fb_r896","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fsam2_hiera_small.fb_r896&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/sam2_hiera_small.fb_r896","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":23.2767,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":24.0983,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":43,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.91,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.39,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"1646327c56a365dc","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/sam2_hiera_tiny.fb_r896","title":"SAM 2 Hiera Tiny","source":"timm/sam2_hiera_tiny.fb_r896","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fsam2_hiera_tiny.fb_r896&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/sam2_hiera_tiny.fb_r896","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":22.2791,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":22.9791,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":44.9,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.83,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.31,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"0f3b74646c9b88fa","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"facebook/sam3","title":"SAM 3","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=facebook%2Fsam3&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"facebook/sam3","hardware":"orin","precision":"INT8","metrics":[{"title":"Latency","subtitle":"NVIDIA Jetson AGX Orin, TensorRT 10.16.2.10, batch 1","metrics":[{"description":"Engine, one photo and prompt (GPU)","value":282.25,"unit":"ms","higherIsBetter":false,"digits":0,"headline":true},{"description":"Photo and prompt to masks (serve.py, median)","value":339.3,"unit":"ms","higherIsBetter":false,"digits":0}]},{"title":"Accuracy","subtitle":"COCO val2017, 300 photos, 893 prompts (each category present, by name)","metrics":[{"description":"Mask AP","value":54.52,"unit":"%","digits":1,"baseline":55.12,"baselineLabel":"FP32","headline":true},{"description":"Mask AP50","value":79.68,"unit":"%","digits":1,"baseline":79.97,"baselineLabel":"FP32"},{"description":"Mask AP75","value":59.3,"unit":"%","digits":1,"baseline":60.22,"baselineLabel":"FP32"}]},{"title":"Memory","metrics":[{"description":"Engine","value":927,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":3178,"baselineLabel":"FP32 weights","improvementWord":"smaller","headline":true},{"description":"serve.py's peak resident memory","value":2316.9,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image","Segmentation"],"build":"2026102722576599","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"facebook/sam3","hardware":"orin-nx-16gb","precision":"INT8","metrics":[{"title":"Latency","subtitle":"NVIDIA Jetson Orin NX 16GB, TensorRT 10.16.2.10, batch 1","metrics":[{"description":"Engine, one photo and prompt (GPU)","value":706.65,"unit":"ms","higherIsBetter":false,"digits":0,"headline":true},{"description":"Photo and prompt to masks (serve.py, median)","value":782.5,"unit":"ms","higherIsBetter":false,"digits":0}]},{"title":"Accuracy","subtitle":"COCO val2017, 300 photos, 893 prompts (each category present, by name)","metrics":[{"description":"Mask AP","value":54.63,"unit":"%","digits":1,"baseline":55.12,"baselineLabel":"FP32","headline":true},{"description":"Mask AP50","value":79.22,"unit":"%","digits":1,"baseline":79.97,"baselineLabel":"FP32"},{"description":"Mask AP75","value":59.32,"unit":"%","digits":1,"baseline":60.22,"baselineLabel":"FP32"}]},{"title":"Memory","metrics":[{"description":"Engine","value":928,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":3178,"baselineLabel":"FP32 weights","improvementWord":"smaller","headline":true},{"description":"serve.py's peak resident memory","value":2270.1,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image","Segmentation"],"build":"2026102809093fd7","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"facebook/sam3","hardware":"orin-nano-super","precision":"INT8","metrics":[{"title":"Latency","subtitle":"NVIDIA Jetson Orin Nano Super, TensorRT 10.16.2.10, batch 1","metrics":[{"description":"Engine, one photo and prompt (GPU)","value":751.22,"unit":"ms","higherIsBetter":false,"digits":0,"headline":true},{"description":"Photo and prompt to masks (serve.py, median)","value":818.5,"unit":"ms","higherIsBetter":false,"digits":0}]},{"title":"Accuracy","subtitle":"COCO val2017, 300 photos, 893 prompts (each category present, by name)","metrics":[{"description":"Mask AP","value":54.63,"unit":"%","digits":1,"baseline":55.12,"baselineLabel":"FP32","headline":true},{"description":"Mask AP50","value":79.22,"unit":"%","digits":1,"baseline":79.97,"baselineLabel":"FP32"},{"description":"Mask AP75","value":59.32,"unit":"%","digits":1,"baseline":60.22,"baselineLabel":"FP32"}]},{"title":"Memory","metrics":[{"description":"Engine","value":928,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":3178,"baselineLabel":"FP32 weights","improvementWord":"smaller","headline":true},{"description":"serve.py's peak resident memory","value":2209.7,"unit":"MiB","higherIsBetter":false,"digits":0}]}],"labels":["Image","Segmentation"],"build":"2026102809097cef","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"facebook/sam3","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"NVIDIA L4, TensorRT 10.3.0, batch 1","metrics":[{"description":"Engine, one photo and prompt (GPU)","value":99.43,"unit":"ms","higherIsBetter":false,"digits":1,"headline":true},{"description":"Photo and prompt to masks (serve.py, median)","value":125.4,"unit":"ms","higherIsBetter":false,"digits":0}]},{"title":"Accuracy","subtitle":"COCO val2017, 300 photos, 893 prompts (each category present, by name)","metrics":[{"description":"Mask AP","value":53.38,"unit":"%","digits":1,"baseline":55.12,"baselineLabel":"FP32","headline":true},{"description":"Mask AP50","value":78.56,"unit":"%","digits":1,"baseline":79.97,"baselineLabel":"FP32"},{"description":"Mask AP75","value":58.35,"unit":"%","digits":1,"baseline":60.22,"baselineLabel":"FP32"}]},{"title":"Size","metrics":[{"description":"Engine","value":929,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":3178,"baselineLabel":"FP32 weights","improvementWord":"smaller","headline":true}]}],"labels":["Image","Segmentation"],"build":"2026101723245c26","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"facebook/sam3","hardware":"rtx-2080-ti","precision":"INT8","metrics":[{"title":"Latency","subtitle":"NVIDIA GeForce RTX 2080 Ti, TensorRT 10.3.0, batch 1","metrics":[{"description":"Engine, one photo and prompt (GPU)","value":92.79,"unit":"ms","higherIsBetter":false,"digits":1,"headline":true},{"description":"Photo and prompt to masks (serve.py, median)","value":127.3,"unit":"ms","higherIsBetter":false,"digits":0}]},{"title":"Accuracy","subtitle":"COCO val2017, 300 photos, 893 prompts (each category present, by name)","metrics":[{"description":"Mask AP","value":53.23,"unit":"%","digits":1,"baseline":55.12,"baselineLabel":"FP32","headline":true},{"description":"Mask AP50","value":78.35,"unit":"%","digits":1,"baseline":79.97,"baselineLabel":"FP32"},{"description":"Mask AP75","value":58.38,"unit":"%","digits":1,"baseline":60.22,"baselineLabel":"FP32"}]},{"title":"Memory","metrics":[{"description":"Engine","value":924,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":3178,"baselineLabel":"FP32 weights","improvementWord":"smaller","headline":true}]}],"labels":["Image","Segmentation"],"build":"202610210611f4fc","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"embedl/sam-3d-body","upstream":"facebook/sam-3d-body-dinov3","title":"SAM 3D Body","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=facebook%2Fsam-3d-body-dinov3&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"embedl/sam-3d-body","hardware":"thor","precision":"FP16","metrics":[{"title":"Latency","subtitle":"batch 1 · one photo, one person, end to end","metrics":[{"description":"Photo to 3D body, one person (median)","value":67.8,"unit":"ms","higherIsBetter":false,"digits":0,"headline":true},{"description":"Encoder, per person (GPU)","value":21.395,"unit":"ms","higherIsBetter":false,"digits":1},{"description":"Decoder + MHR head, per person (GPU)","value":14.146,"unit":"ms","higherIsBetter":false,"digits":1},{"description":"Person detector (GPU)","value":0.966,"unit":"ms","higherIsBetter":false,"digits":2}]},{"title":"Accuracy","subtitle":"mesh error vs the FP32 model · 11 people in 7 COCO photos","metrics":[{"description":"Vertex error vs FP32, mean","value":1.45,"unit":"mm","higherIsBetter":false,"digits":1,"headline":true},{"description":"Vertex error vs FP32, worst person","value":1.87,"unit":"mm","higherIsBetter":false,"digits":1},{"description":"Keypoint error vs FP32 (3D), mean","value":1.02,"unit":"mm","higherIsBetter":false,"digits":1},{"description":"Keypoint error vs FP32 in the photo, mean","value":0.23,"unit":"px","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Engines (detector, encoder, decoder)","value":2078.3,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]}],"labels":["Image","Pose"],"build":"2026101720199dd3","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"embedl/sam-3d-body","hardware":"thor","variant":"huggingface_fp16_tensorrt_tensorrt","precision":"FP16","metrics":[{"title":"Latency","subtitle":"batch 1 · one photo, one person, end to end","metrics":[{"description":"Photo to 3D body, one person (median)","value":67.8,"unit":"ms","higherIsBetter":false,"digits":0,"headline":true},{"description":"Encoder, per person (GPU)","value":21.395,"unit":"ms","higherIsBetter":false,"digits":1},{"description":"Decoder + MHR head, per person (GPU)","value":14.146,"unit":"ms","higherIsBetter":false,"digits":1},{"description":"Person detector (GPU)","value":0.966,"unit":"ms","higherIsBetter":false,"digits":2}]},{"title":"Accuracy","subtitle":"mesh error vs the FP32 model · 11 people in 7 COCO photos","metrics":[{"description":"Vertex error vs FP32, mean","value":1.45,"unit":"mm","higherIsBetter":false,"digits":1,"headline":true},{"description":"Vertex error vs FP32, worst person","value":1.87,"unit":"mm","higherIsBetter":false,"digits":1},{"description":"Keypoint error vs FP32 (3D), mean","value":1.02,"unit":"mm","higherIsBetter":false,"digits":1},{"description":"Keypoint error vs FP32 in the photo, mean","value":0.23,"unit":"px","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Engines (detector, encoder, decoder)","value":2078.3,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]}],"labels":["Image","Pose"],"build":"2026101720199dd3","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"embedl/sam-3d-body","hardware":"l4","precision":"FP16","metrics":[{"title":"Latency","subtitle":"batch 1 · one photo, one person, end to end","metrics":[{"description":"Photo to 3D body, one person (median)","value":92.3,"unit":"ms","higherIsBetter":false,"digits":0,"headline":true},{"description":"Encoder, per person (GPU)","value":45.195,"unit":"ms","higherIsBetter":false,"digits":1},{"description":"Decoder + MHR head, per person (GPU)","value":16.179,"unit":"ms","higherIsBetter":false,"digits":1},{"description":"Person detector (GPU)","value":1.161,"unit":"ms","higherIsBetter":false,"digits":2}]},{"title":"Accuracy","subtitle":"mesh error vs the FP32 model · 11 people in 7 COCO photos","metrics":[{"description":"Vertex error vs FP32, mean","value":1.4,"unit":"mm","higherIsBetter":false,"digits":1,"headline":true},{"description":"Vertex error vs FP32, worst person","value":1.55,"unit":"mm","higherIsBetter":false,"digits":1},{"description":"Keypoint error vs FP32 (3D), mean","value":0.97,"unit":"mm","higherIsBetter":false,"digits":1},{"description":"Keypoint error vs FP32 in the photo, mean","value":0.22,"unit":"px","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Engines (detector, encoder, decoder)","value":2103.9,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]}],"labels":["Image","Pose"],"build":"2026101712534415","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"embedl/sam-3d-body","hardware":"l4","variant":"huggingface_fp16_tensorrt_tensorrt","precision":"FP16","metrics":[{"title":"Latency","subtitle":"batch 1 · one photo, one person, end to end","metrics":[{"description":"Photo to 3D body, one person (median)","value":92.3,"unit":"ms","higherIsBetter":false,"digits":0,"headline":true},{"description":"Encoder, per person (GPU)","value":45.195,"unit":"ms","higherIsBetter":false,"digits":1},{"description":"Decoder + MHR head, per person (GPU)","value":16.179,"unit":"ms","higherIsBetter":false,"digits":1},{"description":"Person detector (GPU)","value":1.161,"unit":"ms","higherIsBetter":false,"digits":2}]},{"title":"Accuracy","subtitle":"mesh error vs the FP32 model · 11 people in 7 COCO photos","metrics":[{"description":"Vertex error vs FP32, mean","value":1.4,"unit":"mm","higherIsBetter":false,"digits":1,"headline":true},{"description":"Vertex error vs FP32, worst person","value":1.55,"unit":"mm","higherIsBetter":false,"digits":1},{"description":"Keypoint error vs FP32 (3D), mean","value":0.97,"unit":"mm","higherIsBetter":false,"digits":1},{"description":"Keypoint error vs FP32 in the photo, mean","value":0.22,"unit":"px","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Engines (detector, encoder, decoder)","value":2103.9,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]}],"labels":["Image","Pose"],"build":"2026101712534415","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"embedl/sam-3d-body","hardware":"l4","variant":"huggingface_int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · one photo, one person, end to end · Baseline TensorRT as baseline","metrics":[{"description":"Photo to 3D body, one person (median)","value":81.6,"unit":"ms","higherIsBetter":false,"digits":0,"baseline":92.3,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Encoder, per person (GPU)","value":31.723,"unit":"ms","higherIsBetter":false,"digits":1,"baseline":45.195,"baselineLabel":"Baseline TensorRT"},{"description":"Decoder + MHR head, per person (GPU)","value":15.958,"unit":"ms","higherIsBetter":false,"digits":1,"baseline":16.179,"baselineLabel":"Baseline TensorRT"},{"description":"Person detector (GPU)","value":1.167,"unit":"ms","higherIsBetter":false,"digits":2,"baseline":1.161,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"mesh error vs the FP32 model · 11 people in 7 COCO photos","metrics":[{"description":"Vertex error vs FP32, mean","value":2.08,"unit":"mm","higherIsBetter":false,"digits":1,"headline":true},{"description":"Vertex error vs FP32, worst person","value":3.51,"unit":"mm","higherIsBetter":false,"digits":1},{"description":"Keypoint error vs FP32 (3D), mean","value":2.26,"unit":"mm","higherIsBetter":false,"digits":1},{"description":"Keypoint error vs FP32 in the photo, mean","value":0.37,"unit":"px","higherIsBetter":false,"digits":2},{"description":"Vertex error vs the same graph in onnxruntime (the compile), mean","value":2.02,"unit":"mm","higherIsBetter":false,"digits":1}]},{"title":"Size","metrics":[{"description":"Engines (detector, encoder, decoder)","value":1533.3,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]}],"labels":["Image","Pose"],"build":"2026101719316ca2","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"embedl/sam-3d-body","hardware":"h200","precision":"FP16","metrics":[{"title":"Latency","subtitle":"batch 1 · one photo, one person, end to end","metrics":[{"description":"Photo to 3D body, one person (median)","value":58.8,"unit":"ms","higherIsBetter":false,"digits":0,"headline":true},{"description":"Encoder, per person (GPU)","value":5.547,"unit":"ms","higherIsBetter":false,"digits":1},{"description":"Decoder + MHR head, per person (GPU)","value":6.59,"unit":"ms","higherIsBetter":false,"digits":1},{"description":"Person detector (GPU)","value":0.712,"unit":"ms","higherIsBetter":false,"digits":2}]},{"title":"Accuracy","subtitle":"mesh error vs the FP32 model · 11 people in 7 COCO photos","metrics":[{"description":"Vertex error vs FP32, mean","value":1.42,"unit":"mm","higherIsBetter":false,"digits":1,"headline":true},{"description":"Vertex error vs FP32, worst person","value":1.74,"unit":"mm","higherIsBetter":false,"digits":1},{"description":"Keypoint error vs FP32 (3D), mean","value":1.05,"unit":"mm","higherIsBetter":false,"digits":1},{"description":"Keypoint error vs FP32 in the photo, mean","value":0.25,"unit":"px","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Engines (detector, encoder, decoder)","value":2108,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]}],"labels":["Image","Pose"],"build":"2026101720190327","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"embedl/sam-3d-body","hardware":"h200","variant":"huggingface_fp16_tensorrt_tensorrt","precision":"FP16","metrics":[{"title":"Latency","subtitle":"batch 1 · one photo, one person, end to end","metrics":[{"description":"Photo to 3D body, one person (median)","value":58.8,"unit":"ms","higherIsBetter":false,"digits":0,"headline":true},{"description":"Encoder, per person (GPU)","value":5.547,"unit":"ms","higherIsBetter":false,"digits":1},{"description":"Decoder + MHR head, per person (GPU)","value":6.59,"unit":"ms","higherIsBetter":false,"digits":1},{"description":"Person detector (GPU)","value":0.712,"unit":"ms","higherIsBetter":false,"digits":2}]},{"title":"Accuracy","subtitle":"mesh error vs the FP32 model · 11 people in 7 COCO photos","metrics":[{"description":"Vertex error vs FP32, mean","value":1.42,"unit":"mm","higherIsBetter":false,"digits":1,"headline":true},{"description":"Vertex error vs FP32, worst person","value":1.74,"unit":"mm","higherIsBetter":false,"digits":1},{"description":"Keypoint error vs FP32 (3D), mean","value":1.05,"unit":"mm","higherIsBetter":false,"digits":1},{"description":"Keypoint error vs FP32 in the photo, mean","value":0.25,"unit":"px","higherIsBetter":false,"digits":2}]},{"title":"Size","metrics":[{"description":"Engines (detector, encoder, decoder)","value":2108,"unit":"MiB","higherIsBetter":false,"digits":0,"headline":true}]}],"labels":["Image","Pose"],"build":"2026101720190327","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/samvit_huge_patch16.sa1b","title":"SAM ViT Huge/16","source":"timm/samvit_huge_patch16.sa1b","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fsamvit_huge_patch16.sa1b&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/samvit_huge_patch16.sa1b","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":253.386,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":264.625,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3.9,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":7.67,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":7.14,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"530328c081133fba","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/sebotnet33ts_256.a1h_in1k","title":"SE-BoTNet-33-TS (256)","source":"timm/sebotnet33ts_256.a1h_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fsebotnet33ts_256.a1h_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/sebotnet33ts_256.a1h_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.28502,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.29492,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":777.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.67,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.04,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.71,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.19,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"a4f5ef2ef893a3ca","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/legacy_seresnet101.in1k","title":"SE-ResNet-101 (legacy)","source":"timm/legacy_seresnet101.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flegacy_seresnet101.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/legacy_seresnet101.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.40319,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.54041,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":703.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.4,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.71,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"58533346906aa5fc","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnet101.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.37984,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.3877,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":294.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.4,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.71,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b8e095440cb8372f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnet101.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.17085,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.18109,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":161.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.4,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.71,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2e1cc4356a6c536c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnet101.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.50245,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.50488,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":664.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.4,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.71,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.07,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.55,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"4a1cdf9beeb2280d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnet101.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.29893,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.30194,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":434.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.4,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.71,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"746056dd104cf787","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/legacy_seresnet152.in1k","title":"SE-ResNet-152 (legacy)","source":"timm/legacy_seresnet152.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flegacy_seresnet152.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/legacy_seresnet152.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.01171,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.15985,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":493.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.6,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.81,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b1dfff378d490eff","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnet152.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.95119,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.96167,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":200.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.6,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.81,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"269b42ffa699188f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnet152.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.93326,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":8.95044,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":111.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.6,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.81,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"bb9b6fbaabffec30","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnet152.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.1708,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.20349,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":460.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.6,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.81,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.27,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.75,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"0ae3e76834cce7d6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnet152.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.30595,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.31,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":302.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.6,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.81,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"27d53ae53d87c532","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/seresnet152d.ra2_in1k","title":"SE-ResNet-152-D","source":"timm/seresnet152d.ra2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fseresnet152d.ra2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/seresnet152d.ra2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.32128,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.39252,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":430.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.2,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.27,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.75,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"f8eb8b4c94c0c776","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/legacy_seresnet18.in1k","title":"SE-ResNet-18 (legacy)","source":"timm/legacy_seresnet18.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flegacy_seresnet18.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/legacy_seresnet18.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.340261,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.397766,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2845.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":71.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.98,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"22623b5a33ab6dd6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnet18.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.719639,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.724976,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1379.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":71.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.98,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8a5913448374eec7","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnet18.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.14676,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.15411,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":866.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":71.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.98,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"887a33608ee6b6d3","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnet18.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.267289,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.268311,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3719.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":71.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.98,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.65,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.13,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"cd46981721ff9a9c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnet18.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.483052,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.485229,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2065.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":71.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.98,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a54eda61bc658a83","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/seresnet33ts.ra2_in1k","title":"SE-ResNet-33-TS","source":"timm/seresnet33ts.ra2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fseresnet33ts.ra2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/seresnet33ts.ra2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.38937,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.39661,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":718.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.82,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.83,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.81,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.29,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"533c6b5514079346","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/legacy_seresnet34.in1k","title":"SE-ResNet-34 (legacy)","source":"timm/legacy_seresnet34.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flegacy_seresnet34.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/legacy_seresnet34.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.545682,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.626343,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1798.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.18,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.85,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6b33a0bb39d17aeb","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnet34.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.23114,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.23694,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":807.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.18,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.85,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"412ec42aecdae831","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnet34.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.98576,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.99341,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":500.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.18,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.85,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8e18f4e8a919fafb","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnet34.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.503461,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.504517,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1980.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":74.18,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.85,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.77,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.25,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"09bebe9e2f95b724","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnet34.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.911043,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.912842,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1096.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":74.18,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.85,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e9c0fd998b56b01d","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/seresnet50.a1_in1k","title":"SE-ResNet-50","source":"timm/seresnet50.a1_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fseresnet50.a1_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/seresnet50.a1_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.72305,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.725098,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1379.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.73,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.84,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.31,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"7591c088bee73249","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/legacy_seresnet50.in1k","title":"SE-ResNet-50 (legacy)","source":"timm/legacy_seresnet50.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flegacy_seresnet50.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/legacy_seresnet50.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.80701,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.902588,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1216.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.49,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"24c4c2fae1758b68","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnet50.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.81984,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.82654,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":546.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.49,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"36817201501dc812","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnet50.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.33196,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.34265,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":298.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.49,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e795d189a739d746","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnet50.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.657874,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.659546,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1516.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.49,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.84,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.31,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"2221506338c28e03","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnet50.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.17192,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.17627,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":852.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":77.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.49,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6d2c4c14acdcf89f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/seresnext101_32x4d.gluon_in1k","title":"SE-ResNeXt-101 32x4d","source":"timm/seresnext101_32x4d.gluon_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fseresnext101_32x4d.gluon_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/seresnext101_32x4d.gluon_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.48813,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.49084,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":671.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.89,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.78,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.07,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.55,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"515322bd21654779","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/legacy_seresnext101_32x4d.in1k","title":"SE-ResNeXt-101 32x4d (legacy)","source":"timm/legacy_seresnext101_32x4d.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flegacy_seresnext101_32x4d.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/legacy_seresnext101_32x4d.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.51415,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.64728,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":653.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.86,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.38,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"56561d41c30ec7a9","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnext101_32x4d.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.43477,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.44934,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":289.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.86,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.38,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"93931a2ef6c17ff8","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnext101_32x4d.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.36222,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.37378,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":156.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.86,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.38,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b541fdda29d75e00","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnext101_32x4d.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.49084,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.49298,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":670,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.86,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.38,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.07,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.55,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"93702c6b63249ea0","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnext101_32x4d.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.19449,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.19904,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":455.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.86,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.38,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2c914af6154772c2","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/seresnext101_32x8d.ah_in1k","title":"SE-ResNeXt-101 32x8d","source":"timm/seresnext101_32x8d.ah_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fseresnext101_32x8d.ah_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/seresnext101_32x8d.ah_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.75858,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.81421,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":568.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.65,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.58,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.05,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"57195db44eecdb25","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/seresnext101d_32x8d.ah_in1k","title":"SE-ResNeXt-101-D 32x8d","source":"timm/seresnext101d_32x8d.ah_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fseresnext101d_32x8d.ah_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/seresnext101d_32x8d.ah_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.86584,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.86884,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":535.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.62,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.76,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.58,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.05,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"8791a6b6094ed9de","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/legacy_seresnext26_32x4d.in1k","title":"SE-ResNeXt-26 32x4d (legacy)","source":"timm/legacy_seresnext26_32x4d.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flegacy_seresnext26_32x4d.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/legacy_seresnext26_32x4d.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.42368,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.490356,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2289.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.55,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"556ceb399e854a30","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnext26_32x4d.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.992302,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.997192,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1001.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.55,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5a2470b69366313b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnext26_32x4d.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.87815,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.88501,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":529.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.55,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6afd9f65befd60e4","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnext26_32x4d.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.354037,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.35498,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2813.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":75.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.55,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.71,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.19,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"c3df0c2d424e2e61","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnext26_32x4d.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.601294,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.603455,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1659.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.55,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"20ad60f2dad5bafd","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/seresnext26d_32x4d.bt_in1k","title":"SE-ResNeXt-26-D 32x4d","source":"timm/seresnext26d_32x4d.bt_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fseresnext26d_32x4d.bt_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/seresnext26d_32x4d.bt_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.383326,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.385376,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2599.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":74.61,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.05,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.71,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.19,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"a3aa0eac3c546ee2","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/seresnext26t_32x4d.bt_in1k","title":"SE-ResNeXt-26-T 32x4d","source":"timm/seresnext26t_32x4d.bt_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fseresnext26t_32x4d.bt_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/seresnext26t_32x4d.bt_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.38044,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.381653,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2616.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":75.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.43,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.71,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.19,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"a3df1e6dde7e2a83","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/seresnext26ts.ch_in1k","title":"SE-ResNeXt-26-TS","source":"timm/seresnext26ts.ch_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fseresnext26ts.ch_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/seresnext26ts.ch_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.616015,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.619019,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1619.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.2,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.44,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.67,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.14,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"6f4400bd6a7e9cf9","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/seresnext50_32x4d.gluon_in1k","title":"SE-ResNeXt-50 32x4d","source":"timm/seresnext50_32x4d.gluon_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fseresnext50_32x4d.gluon_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/seresnext50_32x4d.gluon_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.685646,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.687439,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1455.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.82,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.72,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.83,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.31,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"71ee71243c6008f5","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/legacy_seresnext50_32x4d.in1k","title":"SE-ResNeXt-50 32x4d (legacy)","source":"timm/legacy_seresnext50_32x4d.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flegacy_seresnext50_32x4d.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/legacy_seresnext50_32x4d.in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.882006,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.972168,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1113.6,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.55,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.09,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"efe064bffee2980b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnext50_32x4d.in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.831,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.8378,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":543,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.55,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.09,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a4144cb4ec91e8dc","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnext50_32x4d.in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.48162,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.52411,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":285.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.55,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.09,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"e05428a9b7a8088d","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnext50_32x4d.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.680412,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.681885,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1466.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.55,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.09,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.83,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.31,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"31030cf177e5f5d4","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_seresnext50_32x4d.in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.09022,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.09265,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":916.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.55,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.09,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"c953c24525dc2811","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/seresnextaa101d_32x8d.ah_in1k","title":"SE-ResNeXt-AA-101-D 32x8d","source":"timm/seresnextaa101d_32x8d.ah_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fseresnextaa101d_32x8d.ah_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/seresnextaa101d_32x8d.ah_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.94109,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.94861,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":514.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.07,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.29,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.58,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.05,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"bdaa2a2ca62a436b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/selecsls42b.in1k","title":"SelecSLS-42-B","source":"timm/selecsls42b.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fselecsls42b.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/selecsls42b.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.330307,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.548796,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.331665,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.551086,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":3011.3,"unit":"qps","digits":1,"baseline":1817.57,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.75,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.16,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.88,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.36,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"a46488ee84115818","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/selecsls60.in1k","title":"SelecSLS-60","source":"timm/selecsls60.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fselecsls60.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/selecsls60.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.394271,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.597309,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.395355,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.599915,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":2526.5,"unit":"qps","digits":1,"baseline":1669.31,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.41,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.54,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.86,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.34,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"3a91fd16f34079d7","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/selecsls60b.in1k","title":"SelecSLS-60-B","source":"timm/selecsls60b.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fselecsls60b.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/selecsls60b.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.426067,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.623693,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.427185,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.625031,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":2339.5,"unit":"qps","digits":1,"baseline":1599.43,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.99,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.89,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.89,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.37,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"5d8bc980aa8ec1e5","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/sequencer2d_s.in1k","title":"Sequencer2D-S","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fsequencer2d_s.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/sequencer2d_s.in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":5.124,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":5.151,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":195.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.28,"unit":"%","headline":true},{"description":"Top-5","value":1.31,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/shufflenet_v2_x0_5","title":"ShuffleNetV2 x0.5","source":"torchvision/shufflenet_v2_x0_5","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fshufflenet_v2_x0_5&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/shufflenet_v2_x0_5","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.47389550134539604,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":2110.1698521319277,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":58.138,"unit":"%","digits":2,"baseline":60.552,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":79.986,"unit":"%","digits":2,"baseline":81.746,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"torchvision/shufflenet_v2_x1_0","title":"ShuffleNetV2 x1.0","source":"torchvision/shufflenet_v2_x1_0","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fshufflenet_v2_x1_0&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/shufflenet_v2_x1_0","hardware":"orin-nano-super","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.24023,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":805.155,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNette validation · 3,925 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":77.12101910828025,"unit":"%","digits":2,"baseline":78.64968152866241,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":94.36942675159236,"unit":"%","digits":2,"baseline":95.10828025477707,"baselineLabel":"FP32 model"}]},{"title":"Size","metrics":[{"description":"Engine size","value":2,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"torchvision/shufflenet_v2_x1_0","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.5010382598266006,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":1995.8555666908157,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":67.908,"unit":"%","digits":2,"baseline":69.362,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":87.37,"unit":"%","digits":2,"baseline":88.316,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"torchvision/shufflenet_v2_x1_5","title":"ShuffleNetV2 x1.5","source":"torchvision/shufflenet_v2_x1_5","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fshufflenet_v2_x1_5&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/shufflenet_v2_x1_5","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.5281834211200476,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":1893.2816896816535,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":70.35,"unit":"%","digits":2,"baseline":72.996,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":89.704,"unit":"%","digits":2,"baseline":91.086,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"torchvision/shufflenet_v2_x2_0","title":"ShuffleNetV2 x2.0","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fshufflenet_v2_x2_0&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/shufflenet_v2_x2_0","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":403.198,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":403.992,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":2470.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":71.57,"unit":"%","headline":true},{"description":"Top-5","value":90.13,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/shufflenet_v2_x2_0","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":403.198,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":403.992,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":2470.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":71.57,"unit":"%","headline":true},{"description":"Top-5","value":90.13,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/shvit_s1.in1k","title":"SHViT-S1","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fshvit_s1.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/shvit_s1.in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":582.947,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":583.374,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":1710.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":70.52,"unit":"%","headline":true},{"description":"Top-5","value":89.62,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/shvit_s1.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.58235,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.582855,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1713.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":70.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.7,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.6,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":78,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"2d6cb7e5f776507f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/shvit_s2.in1k","title":"SHViT-S2","source":"timm/shvit_s2.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fshvit_s2.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/shvit_s2.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.652103,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.652588,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1529.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":72.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.97,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.66,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.14,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"53231ab22638e443","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/shvit_s3.in1k","title":"SHViT-S3","source":"timm/shvit_s3.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fshvit_s3.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/shvit_s3.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.799066,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.799683,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1249.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":75.87,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.57,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.69,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.17,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"5720d2909aa286f9","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/shvit_s4.in1k","title":"SHViT-S4","source":"timm/shvit_s4.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fshvit_s4.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/shvit_s4.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.761146,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.762268,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1311.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.02,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.83,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.72,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.2,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"f007e468cb30128f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/skresnet18.ra_in1k","title":"SK-ResNet-18","source":"timm/skresnet18.ra_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fskresnet18.ra_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/skresnet18.ra_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.70813,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.72076,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":367.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":72.82,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.96,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2026102020123ec6","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/skresnet18.ra_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.481184,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.482666,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2071.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":72.82,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.96,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.66,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.13,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"79695032c1fbcf9a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/skresnet34.ra_in1k","title":"SK-ResNet-34","source":"timm/skresnet34.ra_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fskresnet34.ra_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/skresnet34.ra_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.924587,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.926086,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1079.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.15,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.77,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.25,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"3655744d9e886dc4","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/skresnext50_32x4d.ra_in1k","title":"SK-ResNeXt-50 32x4d","source":"timm/skresnext50_32x4d.ra_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fskresnext50_32x4d.ra_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/skresnext50_32x4d.ra_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.951,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.96863,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":143.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.52,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"202610202012cea5","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/skresnext50_32x4d.ra_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.992344,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.08765,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.994019,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.09149,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1005.9,"unit":"qps","digits":1,"baseline":918.011,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.52,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.83,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.31,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e7cfa67f3d61cfbf","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/squeezenet1_0","title":"SqueezeNet 1.0","source":"torchvision/squeezenet1_0","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fsqueezenet1_0&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/squeezenet1_0","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.39312757551670074,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":2543.7035259754202,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":57.132000000000005,"unit":"%","digits":2,"baseline":58.092,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":79.508,"unit":"%","digits":2,"baseline":80.42,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"torchvision/squeezenet1_1","title":"SqueezeNet 1.1","source":"torchvision/squeezenet1_1","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fsqueezenet1_1&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/squeezenet1_1","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.26607225998304784,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":3758.3775176852805,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":57.718,"unit":"%","digits":2,"baseline":58.178,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":79.99000000000001,"unit":"%","digits":2,"baseline":80.624,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"stabilityai/triposr","upstream":"stabilityai/TripoSR","source":"huggingface","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=stabilityai%2FTripoSR&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"stabilityai/triposr","hardware":"h200","precision":"FP16","metrics":[{"title":"Latency","subtitle":"one object crop · warm, two warm-ups then ten timed requests · Baseline PyTorch as baseline","metrics":[{"description":"Crop to GLB, median","value":76.50083501357585,"unit":"ms","higherIsBetter":false,"baseline":77.77806255035102,"baselineLabel":"Baseline PyTorch","headline":true},{"description":"Crop to GLB, p95","value":78.27791734598577,"unit":"ms","higherIsBetter":false},{"description":"Encoder, GPU only","value":1.0569966063062266,"unit":"ms","higherIsBetter":false},{"description":"Encoder, median","value":2.521540503948927,"unit":"ms","higherIsBetter":false},{"description":"Encoder, p95","value":2.5382638443261385,"unit":"ms","higherIsBetter":false},{"description":"Triplane backbone, median","value":7.534903474152088,"unit":"ms","higherIsBetter":false},{"description":"Triplane backbone, p95","value":7.555756042711437,"unit":"ms","higherIsBetter":false},{"description":"Density grid, median","value":18.706608098000288,"unit":"ms","higherIsBetter":false},{"description":"Density grid, p95","value":18.754890374839306,"unit":"ms","higherIsBetter":false},{"description":"Marching cubes, median","value":29.347611940465868,"unit":"ms","higherIsBetter":false},{"description":"Marching cubes, p95","value":30.118102813139558,"unit":"ms","higherIsBetter":false},{"description":"Vertex colours, median","value":2.1460469579324126,"unit":"ms","higherIsBetter":false},{"description":"Vertex colours, p95","value":2.160264633130282,"unit":"ms","higherIsBetter":false},{"description":"Encode, median","value":7.480142987333238,"unit":"ms","higherIsBetter":false},{"description":"Encode, p95","value":9.675024938769639,"unit":"ms","higherIsBetter":false}]},{"title":"Accuracy","subtitle":"4 synthetic objects · mesh agreement with the PyTorch FP32 model","metrics":[{"description":"Chamfer L1 / reference diagonal, mean","value":0.0014130946998156982,"higherIsBetter":false,"digits":5,"headline":true},{"description":"F-score at 1% reference diagonal, worst","value":1,"digits":4}]},{"title":"Memory","metrics":[{"description":"Serving process GPU memory","value":1766,"unit":"MiB","higherIsBetter":false,"headline":true},{"description":"Serving peak RSS","value":1111.83984375,"unit":"MiB","higherIsBetter":false}]}],"labels":["Image"],"build":"202610200808a234","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"stabilityai/triposr","hardware":"h200","variant":"huggingface_fp16_pytorch","precision":"FP16","metrics":[{"title":"Latency","subtitle":"one object crop · warm, two warm-ups then ten timed requests","metrics":[{"description":"Crop to GLB, median","value":77.77806255035102,"unit":"ms","higherIsBetter":false,"headline":true},{"description":"Encoder, GPU only","value":1.0394608354018093,"unit":"ms","higherIsBetter":false}]},{"title":"Accuracy","subtitle":"4 synthetic objects · mesh agreement with the PyTorch FP32 model","metrics":[{"description":"Chamfer L1 / reference diagonal, mean","value":0.0014121138358368429,"higherIsBetter":false,"digits":5,"headline":true},{"description":"F-score at 1% reference diagonal, worst","value":1,"digits":4}]},{"title":"Memory","metrics":[{"description":"Serving process GPU memory","value":1786,"unit":"MiB","higherIsBetter":false,"headline":true},{"description":"Serving peak RSS","value":3098.53125,"unit":"MiB","higherIsBetter":false}]}],"labels":["Image"],"build":"202610180925d34e","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/starnet_s1.in1k","title":"StarNet-S1","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fstarnet_s1.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/starnet_s1.in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":612.213,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":613.037,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":1629.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":40.76,"unit":"%","headline":true},{"description":"Top-5","value":65.71,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/starnet_s3.in1k","title":"StarNet-S3","source":"timm/starnet_s3.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fstarnet_s3.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/starnet_s3.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.616797,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.617981,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1616.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.07,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.54,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.07,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"36cc5abdc1c4465b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/starnet_s4.in1k","title":"StarNet-S4","source":"timm/starnet_s4.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fstarnet_s4.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/starnet_s4.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.820362,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.821838,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1216.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.77,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.26,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.61,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.09,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"95f19fc325235b8b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swiftformer_l1.dist_in1k","title":"SwiftFormer-L1","source":"timm/swiftformer_l1.dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswiftformer_l1.dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swiftformer_l1.dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.833339,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.837891,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1197.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.45,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.14,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.68,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.16,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"2ee303d2c6b3d95e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swiftformer_s.dist_in1k","title":"SwiftFormer-S","source":"timm/swiftformer_s.dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswiftformer_s.dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swiftformer_s.dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.762464,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.765991,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1308.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.33,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.6,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.08,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"a59c4ecd10f7d282","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swiftformer_xs.dist_in1k","title":"SwiftFormer-XS","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswiftformer_xs.dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swiftformer_xs.dist_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":493.896,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":496.582,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":2018,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":59.56,"unit":"%","headline":true},{"description":"Top-5","value":81.35,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swin_s3_base_224.ms_in1k","title":"Swin Base (224, s3)","source":"timm/swin_s3_base_224.ms_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswin_s3_base_224.ms_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swin_s3_base_224.ms_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.56695,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":8.59033,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":116.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.44,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"20261020201280f5","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/swin_s3_base_224.ms_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.88869,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.8952,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":529,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.44,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.32,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.8,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"819f22e2a03b878c","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swin_base_patch4_window7_224.ms_in1k","title":"Swin Base/4 (224, window 7)","source":"timm/swin_base_patch4_window7_224.ms_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswin_base_patch4_window7_224.ms_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swin_base_patch4_window7_224.ms_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.30703,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":7.33447,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":136.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.39,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.33,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2026102020128ead","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/swin_base_patch4_window7_224.ms_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.56561,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.80327,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.57098,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.80646,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":638.1,"unit":"qps","digits":1,"baseline":554.066,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.39,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.33,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"bfa84354536c859e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swin_base_patch4_window12_384.ms_in1k","title":"Swin Base/4 (384, window 12)","source":"timm/swin_base_patch4_window12_384.ms_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswin_base_patch4_window12_384.ms_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swin_base_patch4_window12_384.ms_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":17.6776,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":17.7639,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":56.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":84.21,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.8,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2026102020127571","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/swin_base_patch4_window12_384.ms_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.58377,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.65845,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":278.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.21,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.8,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"41e7b6f80fd4f1ea","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swin_large_patch4_window7_224.ms_in22k_ft_in1k","title":"Swin Large/4 (224, window 7)","source":"timm/swin_large_patch4_window7_224.ms_in22k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswin_large_patch4_window7_224.ms_in22k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swin_large_patch4_window7_224.ms_in22k_ft_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":11.8245,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":11.855,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":84.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":86.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.89,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2026102020120aa7","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/swin_large_patch4_window7_224.ms_in22k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.47123,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.12433,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":2.53528,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.18927,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":404.4,"unit":"qps","digits":1,"baseline":319.917,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":86.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.89,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.72,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.2,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"df68ca6d15614543","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swin_large_patch4_window12_384.ms_in22k_ft_in1k","title":"Swin Large/4 (384, window 12)","source":"timm/swin_large_patch4_window12_384.ms_in22k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswin_large_patch4_window12_384.ms_in22k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swin_large_patch4_window12_384.ms_in22k_ft_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":28.5347,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":28.5903,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":34.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":87.06,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.23,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"202610202012aa74","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/swin_large_patch4_window12_384.ms_in22k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.08824,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.31274,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":164.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":87.06,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.23,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.73,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.21,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"adb1d55a2753ee9e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swin_s3_small_224.ms_in1k","title":"Swin Small (224, s3)","source":"timm/swin_s3_small_224.ms_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswin_s3_small_224.ms_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swin_s3_small_224.ms_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.1411,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.16125,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":162,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":83.1,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.09,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"202610202012e7ed","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/swin_s3_small_224.ms_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.40404,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.40802,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":711.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.1,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.09,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.08,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.56,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"57d83ebb5472b5a4","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swin_small_patch4_window7_224.ms_in1k","title":"Swin Small/4 (224, window 7)","source":"timm/swin_small_patch4_window7_224.ms_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswin_small_patch4_window7_224.ms_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swin_small_patch4_window7_224.ms_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.89169,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.90967,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":168.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.94,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.23,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"202610202012153b","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/swin_small_patch4_window7_224.ms_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.23667,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.24109,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":807.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.94,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.23,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.08,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.56,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"64b3669cfa97d39f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swin_s3_tiny_224.ms_in1k","title":"Swin Tiny (224, s3)","source":"timm/swin_s3_tiny_224.ms_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswin_s3_tiny_224.ms_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swin_s3_tiny_224.ms_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.37282,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.3866,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":294.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.77,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.69,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2026102020125a90","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/swin_s3_tiny_224.ms_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.635405,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.736909,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.637878,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.739624,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1570,"unit":"qps","digits":1,"baseline":1354.5,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.77,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.69,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.84,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.32,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"f6ac7b1c04a235b2","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swin_tiny_patch4_window7_224.ms_in1k","title":"Swin Tiny/4 (224, window 7)","source":"timm/swin_tiny_patch4_window7_224.ms_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswin_tiny_patch4_window7_224.ms_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swin_tiny_patch4_window7_224.ms_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.810395,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.885651,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1199.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.21,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.36,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d19d548cb6bf159f","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/swin_tiny_patch4_window7_224.ms_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.63535,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.64343,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":607.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.21,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.36,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4b027d70563c2786","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/swin_tiny_patch4_window7_224.ms_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.32818,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.33936,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":298.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.21,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.36,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"058c8d71e6493b86","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/swin_tiny_patch4_window7_224.ms_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.580097,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.732573,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.581787,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.734863,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1719.4,"unit":"qps","digits":1,"baseline":1362.45,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.21,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.36,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.84,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.32,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"9049360137e505ec","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/swin_tiny_patch4_window7_224.ms_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.906927,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.911255,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1101.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":81.21,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.36,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7972c2000287342f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swinv2_base_window12_192.ms_in22k","title":"Swin V2 Base (192, window 12)","source":"timm/swinv2_base_window12_192.ms_in22k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswinv2_base_window12_192.ms_in22k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swinv2_base_window12_192.ms_in22k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.56172,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.62122,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":390.1,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.55,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.03,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"5fc41db52fc96604","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swinv2_base_window16_256.ms_in1k","title":"Swin V2 Base (256, window 16)","source":"timm/swinv2_base_window16_256.ms_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswinv2_base_window16_256.ms_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swinv2_base_window16_256.ms_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.2222,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.29431,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":236.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.1,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.56,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.04,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"627784b1fb1ba54c","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swinv2_base_window8_256.ms_in1k","title":"Swin V2 Base (256, window 8)","source":"timm/swinv2_base_window8_256.ms_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswinv2_base_window8_256.ms_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swinv2_base_window8_256.ms_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.55783,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.6358,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":390.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.12,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.91,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.56,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.04,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"9167a8e1f78ff71d","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swinv2_base_window12to16_192to256.ms_in22k_ft_in1k","title":"Swin V2 Base (window12to16, 192to256)","source":"timm/swinv2_base_window12to16_192to256.ms_in22k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswinv2_base_window12to16_192to256.ms_in22k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swinv2_base_window12to16_192to256.ms_in22k_ft_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":15.2066,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":15.294,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":65.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":86.07,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.85,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"202610202012b8d7","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/swinv2_base_window12to16_192to256.ms_in22k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.24717,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.31738,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":235.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":86.07,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.85,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.57,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.04,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e6d7f6b1b500f683","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swinv2_base_window12to24_192to384.ms_in22k_ft_in1k","title":"Swin V2 Base (window12to24, 192to384)","source":"timm/swinv2_base_window12to24_192to384.ms_in22k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswinv2_base_window12to24_192to384.ms_in22k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swinv2_base_window12to24_192to384.ms_in22k_ft_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":42.7439,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":43.0342,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":23.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":86.98,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.21,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"2026102020125491","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/swinv2_base_window12to24_192to384.ms_in22k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":17.4554,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":17.7167,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":57.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":86.98,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.21,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.57,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.05,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"0b73e2d27df86544","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swinv2_large_window12_192.ms_in22k","title":"Swin V2 Large (192, window 12)","source":"timm/swinv2_large_window12_192.ms_in22k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswinv2_large_window12_192.ms_in22k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swinv2_large_window12_192.ms_in22k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.1779,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.24731,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":239.3,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.84,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.31,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"aea1c7cc840179a3","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swinv2_large_window12to16_192to256.ms_in22k_ft_in1k","title":"Swin V2 Large (window12to16, 192to256)","source":"timm/swinv2_large_window12to16_192to256.ms_in22k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswinv2_large_window12to16_192to256.ms_in22k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swinv2_large_window12to16_192to256.ms_in22k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.65925,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.85541,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":176.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":86.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.03,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.85,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.33,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"ea96e63b811c8ad4","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swinv2_large_window12to24_192to384.ms_in22k_ft_in1k","title":"Swin V2 Large (window12to24, 192to384)","source":"timm/swinv2_large_window12to24_192to384.ms_in22k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswinv2_large_window12to24_192to384.ms_in22k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swinv2_large_window12to24_192to384.ms_in22k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":29.9836,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":30.4337,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":33.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":87.43,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.31,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.86,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.34,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"b5017f1120fb5387","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swinv2_cr_small_ns_224.sw_in1k","title":"Swin V2 Small (224, cr, ns)","source":"timm/swinv2_cr_small_ns_224.sw_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswinv2_cr_small_ns_224.sw_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swinv2_cr_small_ns_224.sw_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.70379,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.71588,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":586.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.09,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.38,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.08,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.56,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"fcf5fd2aa8c7b65d","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swinv2_cr_small_224.sw_in1k","title":"Swin V2 Small (224, cr)","source":"timm/swinv2_cr_small_224.sw_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswinv2_cr_small_224.sw_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swinv2_cr_small_224.sw_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.70158,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.71545,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":587.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.02,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.14,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.08,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.56,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"f81c52a43861f944","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swinv2_small_window16_256.ms_in1k","title":"Swin V2 Small (256, window 16)","source":"timm/swinv2_small_window16_256.ms_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswinv2_small_window16_256.ms_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swinv2_small_window16_256.ms_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.1176,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.1759,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":320.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.1,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.8,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.11,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"8a7f324eec9edab3","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swinv2_small_window8_256.ms_in1k","title":"Swin V2 Small (256, window 8)","source":"timm/swinv2_small_window8_256.ms_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswinv2_small_window8_256.ms_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swinv2_small_window8_256.ms_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.9295,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.94745,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":517.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.64,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.59,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.11,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"26020c4c444f05d3","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swinv2_cr_tiny_ns_224.sw_in1k","title":"Swin V2 Tiny (224, cr, ns)","source":"timm/swinv2_cr_tiny_ns_224.sw_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswinv2_cr_tiny_ns_224.sw_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swinv2_cr_tiny_ns_224.sw_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.921095,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.927063,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1083.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.32,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.67,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.84,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.32,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"4c6bfed481cb05e5","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swinv2_tiny_window16_256.ms_in1k","title":"Swin V2 Tiny (256, window 16)","source":"timm/swinv2_tiny_window16_256.ms_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswinv2_tiny_window16_256.ms_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swinv2_tiny_window16_256.ms_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.70428,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.71667,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":586.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.35,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.01,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.86,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.34,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"9612f25101b63a19","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/swinv2_tiny_window8_256.ms_in1k","title":"Swin V2 Tiny (256, window 8)","source":"timm/swinv2_tiny_window8_256.ms_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fswinv2_tiny_window8_256.ms_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/swinv2_tiny_window8_256.ms_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.13351,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.13641,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":881,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.45,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.78,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.86,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.34,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"7ed40e2d7e94f5a6","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/swin_v2_b","title":"Swin V2-B","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fswin_v2_b&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/swin_v2_b","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":2.846,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":2.931,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":348.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":58.51,"unit":"%","headline":true},{"description":"Top-5","value":73.71,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/swin_v2_b","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":2.846,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":2.931,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":348.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":58.51,"unit":"%","headline":true},{"description":"Top-5","value":73.71,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/swin_v2_s","title":"Swin V2-S","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fswin_v2_s&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/swin_v2_s","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":2.205,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":2.217,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":453,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":82.15,"unit":"%","headline":true},{"description":"Top-5","value":96.02,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/swin_v2_s","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":2.205,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":2.217,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":453,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":82.15,"unit":"%","headline":true},{"description":"Top-5","value":96.02,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/swin_v2_t","title":"Swin V2-T","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fswin_v2_t&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/swin_v2_t","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.128,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.141,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":885.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":79.42,"unit":"%","headline":true},{"description":"Top-5","value":94.54,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/swin_v2_t","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.128,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.141,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":885.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":79.42,"unit":"%","headline":true},{"description":"Top-5","value":94.54,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/swin_b","title":"Swin-B","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fswin_b&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/swin_b","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":2.389,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":2.467,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":416.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":79.61,"unit":"%","headline":true},{"description":"Top-5","value":95.15,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/swin_b","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":2.389,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":2.467,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":416.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":79.61,"unit":"%","headline":true},{"description":"Top-5","value":95.15,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/swin_s","title":"Swin-S","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fswin_s&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/swin_s","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.86,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.864,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":537.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":81.11,"unit":"%","headline":true},{"description":"Top-5","value":95.64,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/swin_s","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.86,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.864,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":537.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":81.11,"unit":"%","headline":true},{"description":"Top-5","value":95.64,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/swin_t","title":"Swin-T","source":"torchvision/swin_t","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fswin_t&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/swin_t","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.552002,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.742432,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Throughput","value":1822.04,"unit":"qps","digits":1,"baseline":1347.88,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNette validation · 3,925 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":85.6656050955414,"unit":"%","digits":2,"baseline":86.1656050955414,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Top-5","value":98.36942675159236,"unit":"%","digits":2,"baseline":98.5732484076433,"baselineLabel":"Baseline TensorRT"}]},{"title":"Memory","metrics":[{"description":"Peak device memory","value":39,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":60,"baselineLabel":"Baseline TensorRT","improvementWord":"smaller","headline":true}]},{"title":"Size","metrics":[{"description":"Engine size","value":37.1,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":58.1,"baselineLabel":"Baseline TensorRT","improvementWord":"smaller","headline":true}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"microsoft/swin-tiny-patch4-window7-224","title":"Swin-T (HF)","source":"microsoft/swin-tiny-patch4-window7-224","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=microsoft%2Fswin-tiny-patch4-window7-224&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"microsoft/swin-tiny-patch4-window7-224","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.558594,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.73999,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Throughput","value":1789.01,"unit":"qps","digits":1,"baseline":1348.52,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNette validation · 3,925 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":86.1656050955414,"unit":"%","digits":2,"baseline":86.1656050955414,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Top-5","value":99.05732484076432,"unit":"%","digits":2,"baseline":99.05732484076432,"baselineLabel":"Baseline TensorRT"}]},{"title":"Memory","metrics":[{"description":"Peak device memory","value":36,"unit":"MiB","higherIsBetter":false,"digits":0,"baseline":61,"baselineLabel":"Baseline TensorRT","improvementWord":"smaller","headline":true}]},{"title":"Size","metrics":[{"description":"Engine size","value":36.2,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":60.4,"baselineLabel":"Baseline TensorRT","improvementWord":"smaller","headline":true}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/convmixer_768_32.in1k","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fconvmixer_768_32.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/convmixer_768_32.in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":2.106,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":2.118,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":474.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":45.95,"unit":"%","headline":true},{"description":"Top-5","value":67.64,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/cpubone_nano.r224_in1k","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fcpubone_nano.r224_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/cpubone_nano.r224_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":333.679,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":334.534,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":2983.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":72.71,"unit":"%","headline":true},{"description":"Top-5","value":90.61,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/densenetblur121d.ra_in1k","source":"timm/densenetblur121d.ra_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fdensenetblur121d.ra_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/densenetblur121d.ra_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.778567,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.897278,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1273,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.01,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8de7733bac79876f","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/densenetblur121d.ra_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.47867,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.58069,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":180.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.01,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"557e276f535a60b5","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/densenetblur121d.ra_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.84933,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":7.86719,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":126.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.01,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6f20dd240d02ca2a","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/densenetblur121d.ra_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.63812,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.63916,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":610,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.01,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.61,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":92.7,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"9d560fda8a5b6b85","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/densenetblur121d.ra_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.9047,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.91046,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":344.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":76.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.01,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b501554d50c92d41","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/eca_botnext26ts_256.c1_in1k","source":"timm/eca_botnext26ts_256.c1_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Feca_botnext26ts_256.c1_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/eca_botnext26ts_256.c1_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.535583,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.590942,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1803.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.92,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.42,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"49aecb13bf51b079","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/eca_botnext26ts_256.c1_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.66718,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.69385,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":596.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.92,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.42,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4aae2efb382a0e36","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/eca_botnext26ts_256.c1_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.10531,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.12549,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":320.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.92,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.42,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7be7759f784ed0f7","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/eca_botnext26ts_256.c1_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.719726,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.722412,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1386.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.92,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.42,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.67,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.15,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"f565f02beacea79e","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/eca_botnext26ts_256.c1_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.967732,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.970703,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1032.2,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.92,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.42,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"709bf66e5f92e457","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/ecaresnetlight.miil_in1k","source":"timm/ecaresnetlight.miil_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fecaresnetlight.miil_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/ecaresnetlight.miil_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.833533,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.836853,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1197.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.37,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.69,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.86,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.34,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e4f411b509f74175","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/gernet_l.idstcv_in1k","source":"timm/gernet_l.idstcv_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fgernet_l.idstcv_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/gernet_l.idstcv_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.370576,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.549454,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.43219,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.622437,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":2607.6,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1751.99,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.79,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.84,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"68d12e595df6ec97","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/gernet_l.idstcv_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.758171,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.10032,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.761536,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.10577,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1309.8,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":902.982,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.79,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.84,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"f6db4d2126171f85","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/gernet_l.idstcv_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.34309,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.35095,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":740,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.79,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.84,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"11c684340da10c7a","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/gernet_l.idstcv_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.23533,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.527091,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.236084,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.528931,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":4221.7,"unit":"qps","digits":1,"baseline":1891.55,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.79,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.84,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.87,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.35,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"abfc80953f6132a0","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/gernet_l.idstcv_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.514723,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.516113,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1938.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.79,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.84,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"a33d8d4b66cb45c0","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/gernet_m.idstcv_in1k","source":"timm/gernet_m.idstcv_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fgernet_m.idstcv_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/gernet_m.idstcv_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.241935,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.412754,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.294556,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.461182,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":4070.6,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":2305.56,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.9,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"4e1b4177a4caa73d","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/gernet_m.idstcv_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.614374,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.846706,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.618164,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.85144,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":1615.8,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1172.86,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.9,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6a9affe1b8e7f017","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/gernet_m.idstcv_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.09132,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.09552,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":910.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.9,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"df853313b50c8c8b","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/gernet_m.idstcv_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.182795,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.295656,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.183289,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.296936,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":5430.8,"unit":"qps","digits":1,"baseline":3364.46,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.9,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.76,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.24,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"c2717a30725f4d59","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/gernet_m.idstcv_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.39658,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.397766,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2515.3,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":79.83,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.9,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"106cb191104ac7d3","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/gernet_s.idstcv_in1k","source":"timm/gernet_s.idstcv_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fgernet_s.idstcv_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/gernet_s.idstcv_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.217223,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.27594,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":4237.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.42,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"270cdf9fa148f8bb","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/gernet_s.idstcv_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.442393,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.550209,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.446838,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.554077,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":2241.4,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1803.47,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.42,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d6b17c6f5d5d865b","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/gernet_s.idstcv_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.710839,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.715271,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1396.5,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.42,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"6fa6d61716d8327b","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/gernet_s.idstcv_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.141418,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.193539,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.141968,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.194336,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":6986.1,"unit":"qps","digits":1,"baseline":5104.92,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":75.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.42,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.61,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":94.5,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"194f4f05b3642111","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/gernet_s.idstcv_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.296029,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.298096,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3365.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":75.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.42,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d7320e191afcf1b8","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/lambda_resnet26rpt_256.c1_in1k","source":"timm/lambda_resnet26rpt_256.c1_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flambda_resnet26rpt_256.c1_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/lambda_resnet26rpt_256.c1_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.392941,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.487699,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":0.453369,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.550049,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":2494.8,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":1968.24,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.64,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.27,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"5cb8ff14cc16c8d7","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/lambda_resnet26rpt_256.c1_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.01686,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.02899,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":977.1,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.64,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.27,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"d824930daf25d406","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/lambda_resnet26rpt_256.c1_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.97741,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.99792,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":502.8,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.64,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.27,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"078e05d49e48cd15","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/lambda_resnet26rpt_256.c1_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.291952,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.348516,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.293274,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.350311,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":3409.3,"unit":"qps","digits":1,"baseline":2858.04,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.64,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.27,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.65,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.12,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"867d75834e79251f","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/lambda_resnet26rpt_256.c1_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.519316,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.521912,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1921.4,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":78.64,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.27,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"b888efe83847ae5b","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/naflexvit_base_patch16_parfac_gap.e300_s576_in1k","source":"timm/naflexvit_base_patch16_parfac_gap.e300_s576_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fnaflexvit_base_patch16_parfac_gap.e300_s576_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/naflexvit_base_patch16_parfac_gap.e300_s576_in1k","hardware":"thor","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.35482,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.57993,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"GPU compute, p99","value":1.47668,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.65399,"baselineLabel":"Baseline TensorRT"},{"description":"Throughput","value":717.8,"unit":"qps","higherIsBetter":true,"digits":1,"baseline":620.565,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.99,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.12,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"359b854ba3f97901","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/naflexvit_base_patch16_parfac_gap.e300_s576_in1k","hardware":"orin","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.38756,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.39661,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":226.7,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.99,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.12,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"8a5815b002e509ed","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/naflexvit_base_patch16_parfac_gap.e300_s576_in1k","hardware":"orin-nx-16gb","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 10.16.2.10, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":11.3073,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":11.339,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":88,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.99,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.12,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"7d6d315aeb080a8c","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/naflexvit_base_patch16_parfac_gap.e300_s576_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.37486,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.49353,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":228.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.99,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.12,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"ee8f487bbec4d72f","updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"timm/naflexvit_base_patch16_parfac_gap.e300_s576_in1k","hardware":"rtx-2080-ti","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.76194,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.82068,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":209.9,"unit":"qps","higherIsBetter":true,"digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images · measured on the L4 engine of the same INT8 model","metrics":[{"description":"Top-1, INT8 engine","value":82.99,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.12,"unit":"%","digits":2}]}],"labels":["Image","Classification"],"build":"0fac619e47834a53","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/nasnetalarge.tf_in1k","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fnasnetalarge.tf_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/nasnetalarge.tf_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":6.215,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":6.435,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":159.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":23.57,"unit":"%","headline":true},{"description":"Top-5","value":42.09,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/pnasnet5large.tf_in1k","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fpnasnet5large.tf_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/pnasnet5large.tf_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":6.251,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":6.482,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":159.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":0.1,"unit":"%","headline":true},{"description":"Top-5","value":0.62,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/semnasnet_075.rmsp_in1k","source":"timm/semnasnet_075.rmsp_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fsemnasnet_075.rmsp_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/semnasnet_075.rmsp_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.445772,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.446594,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2235.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":72.72,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.03,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.56,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.04,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"110054264440fcaf","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/semnasnet_100.rmsp_in1k","source":"timm/semnasnet_100.rmsp_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fsemnasnet_100.rmsp_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/semnasnet_100.rmsp_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.269877,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.270447,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3684.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":72.63,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.86,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.57,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":46,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"403267a02a914934","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/spnasnet_100.rmsp_in1k","source":"timm/spnasnet_100.rmsp_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fspnasnet_100.rmsp_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/spnasnet_100.rmsp_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.182534,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.204989,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.182922,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.205994,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":5431.8,"unit":"qps","digits":1,"baseline":4839.36,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":72.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.49,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.57,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":52.1,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"c582b67701efe4e9","updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/tinynet_a.in1k","title":"TinyNet-A","source":"timm/tinynet_a.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftinynet_a.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tinynet_a.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.675276,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.677856,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1477.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.35,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.33,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.6,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.08,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"45d67e6d8e639d8f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tinynet_b.in1k","title":"TinyNet-B","source":"timm/tinynet_b.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftinynet_b.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tinynet_b.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.55906,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.560974,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1784.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":74.67,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.04,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.57,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.05,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"274d891c1adbf5de","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tinynet_c.in1k","title":"TinyNet-C","source":"timm/tinynet_c.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftinynet_c.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tinynet_c.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.475642,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.476501,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2096.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":70.79,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.5,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.55,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.03,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"d76d6df5010d1519","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tinynet_d.in1k","title":"TinyNet-D","source":"timm/tinynet_d.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftinynet_d.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tinynet_d.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.321184,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.321777,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3097.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":66.48,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":86.92,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.55,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.03,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"0169808d89dee331","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tiny_vit_21m_224.dist_in22k_ft_in1k","title":"TinyViT 21M (224)","source":"timm/tiny_vit_21m_224.dist_in22k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftiny_vit_21m_224.dist_in22k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tiny_vit_21m_224.dist_in22k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.792939,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.804138,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1259,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.16,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.25,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.76,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.24,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"1dadd77957269c81","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tiny_vit_21m_384.dist_in22k_ft_in1k","title":"TinyViT 21M (384)","source":"timm/tiny_vit_21m_384.dist_in22k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftiny_vit_21m_384.dist_in22k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tiny_vit_21m_384.dist_in22k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.78787,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.89734,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":358.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.19,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.77,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.25,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"f6a793040d06cfd1","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tiny_vit_21m_512.dist_in22k_ft_in1k","title":"TinyViT 21M (512)","source":"timm/tiny_vit_21m_512.dist_in22k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftiny_vit_21m_512.dist_in22k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tiny_vit_21m_512.dist_in22k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.26221,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":9.54517,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":107.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.55,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.4,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.77,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.25,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"1653dd33fc08224c","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tiny_vit_5m_224.dist_in22k_ft_in1k","title":"TinyViT 5M (224)","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftiny_vit_5m_224.dist_in22k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tiny_vit_5m_224.dist_in22k_ft_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":547.974,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":549.622,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":1819.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":25.46,"unit":"%","headline":true},{"description":"Top-5","value":46.89,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tnt_b_patch16_224.in1k","title":"TNT-B/16 (224)","source":"timm/tnt_b_patch16_224.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftnt_b_patch16_224.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tnt_b_patch16_224.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.63919,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.66614,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":609.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.12,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.26,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.74,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"14653d353e162412","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tnt_s_legacy_patch16_224.in1k","title":"TNT-S/16 (224, legacy)","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftnt_s_legacy_patch16_224.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tnt_s_legacy_patch16_224.in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.289,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.292,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":774.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":79.26,"unit":"%","headline":true},{"description":"Top-5","value":94.73,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tnt_s_patch16_224.in1k","title":"TNT-S/16 (224)","source":"timm/tnt_s_patch16_224.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftnt_s_patch16_224.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tnt_s_patch16_224.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.11303,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.307,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.11554,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.31262,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":897.2,"unit":"qps","digits":1,"baseline":764.131,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.89,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.51,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.79,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.27,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"d8863807a776dab4","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/alexnet","source":"torchvision/alexnet","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Falexnet&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/alexnet","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.6069742795079947,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":1647.5162684168213,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":56.232000000000006,"unit":"%","digits":2,"baseline":56.522,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":79.01599999999999,"unit":"%","digits":2,"baseline":79.066,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"torchvision/googlenet","source":"torchvision/googlenet","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fgooglenet&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/googlenet","hardware":"orin-nano-super","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":2.98242,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":334.969,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNette validation · 3,925 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":75.13375796178345,"unit":"%","digits":2,"baseline":75.51592356687898,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":93.5031847133758,"unit":"%","digits":2,"baseline":93.73248407643312,"baselineLabel":"FP32 model"}]},{"title":"Size","metrics":[{"description":"Engine size","value":8,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"torchvision/googlenet","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.9799706799094565,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":1020.4386932192646,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":69.51,"unit":"%","digits":2,"baseline":69.778,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":89.198,"unit":"%","digits":2,"baseline":89.53,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/tresnet_l.miil_in1k","title":"TResNet-L","source":"timm/tresnet_l.miil_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftresnet_l.miil_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tresnet_l.miil_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.28202,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.44293,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.28503,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.44672,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":779.2,"unit":"qps","digits":1,"baseline":692.354,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.47,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.22,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.94,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.42,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"b5174edc7d7eece3","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tresnet_m.miil_in1k","title":"TResNet-M","source":"timm/tresnet_m.miil_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftresnet_m.miil_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tresnet_m.miil_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.753334,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.842406,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.755005,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.846802,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1324.4,"unit":"qps","digits":1,"baseline":1185.12,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.99,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.4,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.76,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.24,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"6cc002dd6c92ec94","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/tresnet_xl.miil_in1k","title":"TResNet-XL","source":"timm/tresnet_xl.miil_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftresnet_xl.miil_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/tresnet_xl.miil_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.58539,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.9738,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.59055,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.99231,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":630.2,"unit":"qps","digits":1,"baseline":506.218,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.14,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.48,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.11,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"64b29d3d5dd0534b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/twins_svt_base.in1k","title":"Twins-SVT Base","source":"timm/twins_svt_base.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftwins_svt_base.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/twins_svt_base.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.14529,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.18512,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":465.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.23,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.55,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.17,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.65,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"a28662a41ae6ce0d","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/twins_svt_large.in1k","title":"Twins-SVT Large","source":"timm/twins_svt_large.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftwins_svt_large.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/twins_svt_large.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.83411,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.93188,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":352.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.31,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.79,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.67,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.15,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"415138b8b9a5b0ce","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/twins_svt_small.in1k","title":"Twins-SVT Small","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Ftwins_svt_small.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/twins_svt_small.in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.072,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.074,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":931.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":77.63,"unit":"%","headline":true},{"description":"Top-5","value":93.63,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/twins_svt_small.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.04686,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.05157,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":953.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.68,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.8,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.28,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"02b0712f78f01daf","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/vgg11","title":"VGG-11","source":"torchvision/vgg11","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvgg11.tv_in1k&granularity=1&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/vgg11","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.7497700955718756,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":1333.7421776434887,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":68.84,"unit":"%","digits":2,"baseline":69.02,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":88.544,"unit":"%","digits":2,"baseline":88.628,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/vgg11.tv_in1k","title":"VGG-11 (timm)","source":"timm/vgg11.tv_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvgg11.tv_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vgg11.tv_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.649713,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.33424,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.6521,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.33673,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1535.3,"unit":"qps","digits":1,"baseline":748.543,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":68.81,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":88.5,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.48,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"5cf1a14410844c4a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/vgg11_bn","title":"VGG-11 BN","source":"torchvision/vgg11_bn","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvgg11_bn.tv_in1k&granularity=1&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/vgg11_bn","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.752695519477129,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":1328.5584597270683,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":69.64399999999999,"unit":"%","digits":2,"baseline":70.37,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":89.266,"unit":"%","digits":2,"baseline":89.81,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/vgg11_bn.tv_in1k","title":"VGG-11 BN (timm)","source":"timm/vgg11_bn.tv_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvgg11_bn.tv_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vgg11_bn.tv_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.653944,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.33002,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.656982,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.33246,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1525.4,"unit":"qps","digits":1,"baseline":751.06,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":69.71,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.26,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.48,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"bcd63d76e48de19d","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/vgg13","title":"VGG-13","source":"torchvision/vgg13","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvgg13.tv_in1k&granularity=1&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/vgg13","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.8044700790196657,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":1243.0543112536998,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":69.804,"unit":"%","digits":2,"baseline":69.928,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":89.218,"unit":"%","digits":2,"baseline":89.246,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/vgg13.tv_in1k","title":"VGG-13 (timm)","source":"timm/vgg13.tv_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvgg13.tv_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vgg13.tv_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.70737,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.46679,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.709412,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.46954,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1410.6,"unit":"qps","digits":1,"baseline":680.983,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":69.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":89.27,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.49,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"b063eb1c839dfccf","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/vgg13_bn","title":"VGG-13 BN","source":"torchvision/vgg13_bn","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvgg13_bn.tv_in1k&granularity=1&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/vgg13_bn","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.8084426820278168,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":1236.946072035311,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":71.20400000000001,"unit":"%","digits":2,"baseline":71.586,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":90.14,"unit":"%","digits":2,"baseline":90.374,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/vgg13_bn.tv_in1k","title":"VGG-13 BN (timm)","source":"timm/vgg13_bn.tv_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvgg13_bn.tv_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vgg13_bn.tv_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.711761,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.46509,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.714355,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.46973,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1401.6,"unit":"qps","digits":1,"baseline":681.878,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":71.19,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.16,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.01,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.48,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"58f948b40d507c16","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/vgg16","title":"VGG-16","source":"torchvision/vgg16","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvgg16.tv_in1k&granularity=1&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/vgg16","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.8705489998101257,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":1148.7004180328836,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":71.474,"unit":"%","digits":2,"baseline":71.592,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":90.302,"unit":"%","digits":2,"baseline":90.382,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/vgg16.tv_in1k","title":"VGG-16 (timm)","source":"timm/vgg16.tv_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvgg16.tv_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vgg16.tv_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.785016,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.61659,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.787964,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.62457,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1271.3,"unit":"qps","digits":1,"baseline":618.04,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":71.41,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.28,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.07,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.54,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"eef50f49758e7137","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/vgg16_bn","title":"VGG-16 BN","source":"torchvision/vgg16_bn","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvgg16_bn.tv_in1k&granularity=1&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/vgg16_bn","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.8762450609356165,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":1141.2332514972961,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":73.08200000000001,"unit":"%","digits":2,"baseline":73.36,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":91.214,"unit":"%","digits":2,"baseline":91.516,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/vgg16_bn.tv_in1k","title":"VGG-16 BN (timm)","source":"timm/vgg16_bn.tv_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvgg16_bn.tv_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vgg16_bn.tv_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.786319,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.60821,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.789734,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.61621,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1269.1,"unit":"qps","digits":1,"baseline":621.254,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":73.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.25,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.06,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.54,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"18321f746e0d88ee","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/vgg19","title":"VGG-19","source":"torchvision/vgg19","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvgg19.tv_in1k&granularity=1&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/vgg19","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.9269990198663436,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":1078.7497921456074,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":72.21600000000001,"unit":"%","digits":2,"baseline":72.376,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":90.778,"unit":"%","digits":2,"baseline":90.876,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/vgg19.tv_in1k","title":"VGG-19 (timm)","source":"timm/vgg19.tv_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvgg19.tv_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vgg19.tv_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.852678,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.75765,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.855408,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.80786,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1170.6,"unit":"qps","digits":1,"baseline":568.471,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":72.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":90.78,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.12,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.6,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e0e21ce2b0674b4e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/vgg19_bn","title":"VGG-19 BN","source":"torchvision/vgg19_bn","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvgg19_bn.tv_in1k&granularity=1&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/vgg19_bn","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.9264079388231039,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":1079.4380726814436,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":73.556,"unit":"%","digits":2,"baseline":74.218,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":91.514,"unit":"%","digits":2,"baseline":91.842,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/vgg19_bn.tv_in1k","title":"VGG-19 BN (timm)","source":"timm/vgg19_bn.tv_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvgg19_bn.tv_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vgg19_bn.tv_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.857526,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.74744,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.861847,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.79291,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1164.2,"unit":"qps","digits":1,"baseline":571.748,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":73.61,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.54,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.12,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.6,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"22b2cec81db6d9a9","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/visformer_small.in1k","title":"Visformer Small","source":"timm/visformer_small.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvisformer_small.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/visformer_small.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.50774,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.51135,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":662.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.2,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.52,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.05,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.53,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"4743299981e36014","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/visformer_tiny.in1k","title":"Visformer Tiny","source":"timm/visformer_tiny.in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvisformer_tiny.in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/visformer_tiny.in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.718531,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.720337,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1388.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.93,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.66,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.14,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"22996d38b0a038f9","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_mci_224.apple_mclip","title":"ViT Base (224, mci)","source":"timm/vit_base_mci_224.apple_mclip","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_mci_224.apple_mclip&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_mci_224.apple_mclip","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.995693,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.29676,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.998474,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.31274,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1002.9,"unit":"qps","digits":1,"baseline":770.273,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.49,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.97,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"3410a0706db72ea0","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_r50_s16_224.orig_in21k","title":"ViT Base (224, r50, s16)","source":"timm/vit_base_r50_s16_224.orig_in21k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_r50_s16_224.orig_in21k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_r50_s16_224.orig_in21k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.96959,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.32836,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":2.0199,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.40375,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":507.3,"unit":"qps","digits":1,"baseline":429.223,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.62,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.1,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"b547d42cd7e2efe1","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch16_plus_clip_240.laion400m_e32","title":"ViT Base Plus/16 (240, CLIP)","source":"timm/vit_base_patch16_plus_clip_240.laion400m_e32","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch16_plus_clip_240.laion400m_e32&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch16_plus_clip_240.laion400m_e32","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.21295,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.7376,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.2196,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.80475,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":823.5,"unit":"qps","digits":1,"baseline":574.992,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.83,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.31,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"1f372c61aabe4b40","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_relpos_base_patch32_plus_rpn_256.sw_in1k","title":"ViT Base Plus/32 (256, relpos, rpn)","source":"timm/vit_relpos_base_patch32_plus_rpn_256.sw_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_relpos_base_patch32_plus_rpn_256.sw_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_relpos_base_patch32_plus_rpn_256.sw_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.01787,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.4293,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.01996,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.4339,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":981,"unit":"qps","digits":1,"baseline":698.792,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.72,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.72,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.86,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.34,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"b72eea90219d8d0a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch14_dinov2.lvd142m","title":"ViT Base/14 (DINOv2)","source":"timm/vit_base_patch14_dinov2.lvd142m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch14_dinov2.lvd142m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch14_dinov2.lvd142m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":25.5821,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":26.4728,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":39.1,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"f56e03b51dfeea2b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch14_reg1_tipsv2.webli","title":"ViT Base/14 (reg1, tipsv2)","source":"timm/vit_base_patch14_reg1_tipsv2.webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch14_reg1_tipsv2.webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch14_reg1_tipsv2.webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":16.19,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":17.2289,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":61.8,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"2733a4b573a4c1b2","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch14_reg4_dinov2.lvd142m","title":"ViT Base/14 (reg4, DINOv2)","source":"timm/vit_base_patch14_reg4_dinov2.lvd142m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch14_reg4_dinov2.lvd142m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch14_reg4_dinov2.lvd142m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":25.6843,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":26.3066,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":38.9,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"8c6a0c6132290bb7","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch16_clip_224.datacompxl","title":"ViT Base/16 (224, CLIP, DataComp-XL)","source":"timm/vit_base_patch16_clip_224.datacompxl","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch16_clip_224.datacompxl&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch16_clip_224.datacompxl","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.972342,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.30534,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.97583,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.30981,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1026.7,"unit":"qps","digits":1,"baseline":765.223,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.49,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.97,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"e30d06c7f755de11","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch16_clip_224.laion2b_ft_in1k","title":"ViT Base/16 (224, CLIP, LAION-2B → IN-1k)","source":"timm/vit_base_patch16_clip_224.laion2b_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch16_clip_224.laion2b_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch16_clip_224.laion2b_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.981775,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.30408,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.984802,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.31281,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1016.9,"unit":"qps","digits":1,"baseline":765.934,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.74,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.57,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.49,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.97,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"30afbd3e65a88d37","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch16_clip_224.openai","title":"ViT Base/16 (224, CLIP, OpenAI)","source":"timm/vit_base_patch16_clip_224.openai","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch16_clip_224.openai&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch16_clip_224.openai","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.98095,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.29461,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.984161,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.29742,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1017.9,"unit":"qps","digits":1,"baseline":771.207,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.3,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.77,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"3c2cf6c056a95452","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch16_224.dino","title":"ViT Base/16 (224, DINO)","source":"timm/vit_base_patch16_224.dino","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch16_224.dino&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch16_224.dino","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.974129,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.28283,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.981384,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.28937,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1024.7,"unit":"qps","digits":1,"baseline":778.524,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.49,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.96,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"4203f478f05593b7","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch16_siglip_gap_224.v2_webli","title":"ViT Base/16 (224, GAP, SigLIP)","source":"timm/vit_base_patch16_siglip_gap_224.v2_webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch16_siglip_gap_224.v2_webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch16_siglip_gap_224.v2_webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.972518,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.36535,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.975037,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.36932,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1026.7,"unit":"qps","digits":1,"baseline":731.59,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.48,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.96,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"3d4688019e53af53","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch16_224_miil.in21k_ft_in1k","title":"ViT Base/16 (224, miil)","source":"timm/vit_base_patch16_224_miil.in21k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch16_224_miil.in21k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch16_224_miil.in21k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.95995,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.31852,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.963074,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.32123,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1040,"unit":"qps","digits":1,"baseline":757.477,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.81,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.49,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.97,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"935f83e33ee16485","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_pe_core_base_patch16_224.fb","title":"ViT Base/16 (224, PE, Core)","source":"timm/vit_pe_core_base_patch16_224.fb","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_pe_core_base_patch16_224.fb&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_pe_core_base_patch16_224.fb","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.62392,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.6452,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":615.2,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.07,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"3e365509666f00d7","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch16_clip_quickgelu_224.openai","title":"ViT Base/16 (224, QuickGELU, CLIP)","source":"timm/vit_base_patch16_clip_quickgelu_224.openai","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch16_clip_quickgelu_224.openai&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch16_clip_quickgelu_224.openai","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.948543,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.31651,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.952759,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.32928,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1052.4,"unit":"qps","digits":1,"baseline":758.627,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.3,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.77,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"8fc44f4e2206a21b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_relpos_base_patch16_clsgap_224.sw_in1k","title":"ViT Base/16 (224, relpos, clsgap)","source":"timm/vit_relpos_base_patch16_clsgap_224.sw_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_relpos_base_patch16_clsgap_224.sw_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_relpos_base_patch16_clsgap_224.sw_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.33404,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.77935,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.33911,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.83624,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":748.7,"unit":"qps","digits":1,"baseline":561.55,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.48,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.52,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.49,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.97,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"7549304d46e91967","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_relpos_base_patch16_224.sw_in1k","title":"ViT Base/16 (224, relpos)","source":"timm/vit_relpos_base_patch16_224.sw_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_relpos_base_patch16_224.sw_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_relpos_base_patch16_224.sw_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.09654,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.53971,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.10107,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.54388,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":910.7,"unit":"qps","digits":1,"baseline":648.853,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.26,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.06,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.49,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.97,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e81e473cf64ca022","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch16_rope_ape_224.naver_in1k","title":"ViT Base/16 (224, RoPE, ape)","source":"timm/vit_base_patch16_rope_ape_224.naver_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch16_rope_ape_224.naver_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch16_rope_ape_224.naver_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.54953,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.56207,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":644.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.61,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.56,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"9bed90565589d233","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch16_rope_224.naver_in1k","title":"ViT Base/16 (224, RoPE)","source":"timm/vit_base_patch16_rope_224.naver_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch16_rope_224.naver_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch16_rope_224.naver_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.55131,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.56311,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":643.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.53,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.59,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"5ac2de0fae7c5e90","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch16_rpn_224.sw_in1k","title":"ViT Base/16 (224, rpn)","source":"timm/vit_base_patch16_rpn_224.sw_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch16_rpn_224.sw_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch16_rpn_224.sw_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.00179,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.29115,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.00623,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.31812,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":996.4,"unit":"qps","digits":1,"baseline":773.601,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.29,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.7,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.49,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.97,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"480c533369a68dfc","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch16_siglip_224.v2_webli","title":"ViT Base/16 (224, SigLIP)","source":"timm/vit_base_patch16_siglip_224.v2_webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch16_siglip_224.v2_webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch16_siglip_224.v2_webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.02025,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.42271,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.02264,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.42822,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":978.8,"unit":"qps","digits":1,"baseline":702.14,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.56,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.04,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"fc5547da808c79df","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch16_224.augreg2_in21k_ft_in1k","title":"ViT Base/16 (224)","source":"timm/vit_base_patch16_224.augreg2_in21k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch16_224.augreg2_in21k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch16_224.augreg2_in21k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.97658,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.28216,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.982727,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.28497,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1022.5,"unit":"qps","digits":1,"baseline":778.828,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.21,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.55,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.49,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.97,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"581ce26eee562637","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch16_rope_reg1_gap_256.sbb_in1k","title":"ViT Base/16 (256, RoPE, reg1, GAP)","source":"timm/vit_base_patch16_rope_reg1_gap_256.sbb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch16_rope_reg1_gap_256.sbb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch16_rope_reg1_gap_256.sbb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.6189,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.67529,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":381.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.77,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.01,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.57,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.04,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"d426fda269629390","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch16_siglip_256.v2_webli","title":"ViT Base/16 (256, SigLIP)","source":"timm/vit_base_patch16_siglip_256.v2_webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch16_siglip_256.v2_webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch16_siglip_256.v2_webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.0756,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.51882,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.08337,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.52252,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":928.3,"unit":"qps","digits":1,"baseline":657.74,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.56,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.04,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"8bf92cd0562f1050","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch16_clip_384.laion2b_ft_in1k","title":"ViT Base/16 (384, CLIP)","source":"timm/vit_base_patch16_clip_384.laion2b_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch16_clip_384.laion2b_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch16_clip_384.laion2b_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.29613,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.52008,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":188.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":85.41,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.67,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"41f409e74fbf3ba7","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch16_siglip_384.v2_webli","title":"ViT Base/16 (384, SigLIP)","source":"timm/vit_base_patch16_siglip_384.v2_webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch16_siglip_384.v2_webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch16_siglip_384.v2_webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.09009,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.2782,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":244.4,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.06,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"0fa62f5123785db2","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch16_384.augreg_in1k","title":"ViT Base/16 (384)","source":"timm/vit_base_patch16_384.augreg_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch16_384.augreg_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch16_384.augreg_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.32053,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.52466,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":187.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.77,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.07,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"c7d70b4447b21cb9","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_pe_spatial_base_patch16_512.fb","title":"ViT Base/16 (512, PE, Spatial)","source":"timm/vit_pe_spatial_base_patch16_512.fb","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_pe_spatial_base_patch16_512.fb&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_pe_spatial_base_patch16_512.fb","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":17.2523,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":17.7639,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":58,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"66252c21a3a4ecf4","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch16_siglip_512.v2_webli","title":"ViT Base/16 (512, SigLIP)","source":"timm/vit_base_patch16_siglip_512.v2_webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch16_siglip_512.v2_webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch16_siglip_512.v2_webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":14.5402,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":14.9592,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":68.8,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.07,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"6e0994438f781680","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch16_dinov3.lvd1689m","title":"ViT Base/16 (DINOv3)","source":"timm/vit_base_patch16_dinov3.lvd1689m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch16_dinov3.lvd1689m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch16_dinov3.lvd1689m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.27615,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.27948,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":782.7,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.48,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.96,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"17d64d7832642e28","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch16_lingbot.robbyant","title":"ViT Base/16 (lingbot)","source":"timm/vit_base_patch16_lingbot.robbyant","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch16_lingbot.robbyant&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch16_lingbot.robbyant","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":18.8027,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":19.2867,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":53.2,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.55,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.03,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"e28dca19f429db84","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch16_dinov3_qkvb.eupe_lvd1689m","title":"ViT Base/16 (qkvb, DINOv3)","source":"timm/vit_base_patch16_dinov3_qkvb.eupe_lvd1689m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch16_dinov3_qkvb.eupe_lvd1689m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch16_dinov3_qkvb.eupe_lvd1689m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.74051,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.74628,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":574,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.53,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.01,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"599396491275b971","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch16_sapiens2.fb","title":"ViT Base/16 (sapiens2)","source":"timm/vit_base_patch16_sapiens2.fb","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch16_sapiens2.fb&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch16_sapiens2.fb","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":144.149,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":149.234,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":6.9,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.09,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.57,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"68b17b8a416da61f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch32_clip_224.laion2b_ft_in1k","title":"ViT Base/32 (224, CLIP, LAION-2B → IN-1k)","source":"timm/vit_base_patch32_clip_224.laion2b_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch32_clip_224.laion2b_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch32_clip_224.laion2b_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.796903,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.00696,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.798767,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.00903,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1252.5,"unit":"qps","digits":1,"baseline":991.714,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.08,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.57,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"ce6e5562fd16b205","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch32_clip_224.openai","title":"ViT Base/32 (224, CLIP, OpenAI)","source":"timm/vit_base_patch32_clip_224.openai","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch32_clip_224.openai&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch32_clip_224.openai","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.794891,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.00284,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.796265,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.00494,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1255.5,"unit":"qps","digits":1,"baseline":995.424,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.31,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.79,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"31b6746a72135550","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch32_clip_quickgelu_224.laion400m_e32","title":"ViT Base/32 (224, QuickGELU, CLIP, LAION-400M, e32)","source":"timm/vit_base_patch32_clip_quickgelu_224.laion400m_e32","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch32_clip_quickgelu_224.laion400m_e32&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch32_clip_quickgelu_224.laion400m_e32","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.780309,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.01162,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.782776,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.01385,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1279.1,"unit":"qps","digits":1,"baseline":987.216,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"449751ff21b844a6","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch32_clip_quickgelu_224.openai","title":"ViT Base/32 (224, QuickGELU, CLIP, OpenAI)","source":"timm/vit_base_patch32_clip_quickgelu_224.openai","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch32_clip_quickgelu_224.openai&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch32_clip_quickgelu_224.openai","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.780968,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.0134,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.78302,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.01514,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1277.6,"unit":"qps","digits":1,"baseline":985.076,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.31,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.79,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"f7db27bebb6967e8","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch32_224.augreg_in1k","title":"ViT Base/32 (224)","source":"timm/vit_base_patch32_224.augreg_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch32_224.augreg_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch32_224.augreg_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.788514,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.00573,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.789978,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.00757,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1266,"unit":"qps","digits":1,"baseline":992.759,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":74.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.32,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"10ad9742a82d69b0","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch32_clip_256.datacompxl","title":"ViT Base/32 (256, CLIP)","source":"timm/vit_base_patch32_clip_256.datacompxl","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch32_clip_256.datacompxl&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch32_clip_256.datacompxl","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.893288,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.11268,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.89502,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.11493,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1117.5,"unit":"qps","digits":1,"baseline":897.517,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"27b344b6f945cb10","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch32_siglip_gap_256.v2_webli","title":"ViT Base/32 (256, GAP, SigLIP)","source":"timm/vit_base_patch32_siglip_gap_256.v2_webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch32_siglip_gap_256.v2_webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch32_siglip_gap_256.v2_webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.78439,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.10487,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.786438,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.10693,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1272,"unit":"qps","digits":1,"baseline":903.961,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.5,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.98,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"cd8536bf00b6cd31","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch32_siglip_256.v2_webli","title":"ViT Base/32 (256, SigLIP)","source":"timm/vit_base_patch32_siglip_256.v2_webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch32_siglip_256.v2_webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch32_siglip_256.v2_webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.811403,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.16161,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.814331,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.1637,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1230.1,"unit":"qps","digits":1,"baseline":859.769,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.58,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.06,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"d1ac13e38bda9883","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch32_clip_384.laion2b_ft_in12k_in1k","title":"ViT Base/32 (384, CLIP)","source":"timm/vit_base_patch32_clip_384.laion2b_ft_in12k_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch32_clip_384.laion2b_ft_in12k_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch32_clip_384.laion2b_ft_in12k_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.06389,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.25298,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.06702,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.25659,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":938.5,"unit":"qps","digits":1,"baseline":797.073,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.8,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"1cc3b52f053030dd","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch32_384.augreg_in1k","title":"ViT Base/32 (384)","source":"timm/vit_base_patch32_384.augreg_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch32_384.augreg_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch32_384.augreg_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.06725,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.25776,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.07086,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.26453,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":935.5,"unit":"qps","digits":1,"baseline":794.138,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.19,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.97,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"bd8cf99440e1c3fd","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch32_clip_448.laion2b_ft_in12k_in1k","title":"ViT Base/32 (448, CLIP)","source":"timm/vit_base_patch32_clip_448.laion2b_ft_in12k_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch32_clip_448.laion2b_ft_in12k_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch32_clip_448.laion2b_ft_in12k_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.18382,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.3886,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.18872,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.39148,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":843.6,"unit":"qps","digits":1,"baseline":719.201,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.96,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.06,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"81c8cccbe69a2b2f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch8_224.dino","title":"ViT Base/8 (224, DINO)","source":"timm/vit_base_patch8_224.dino","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch8_224.dino&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch8_224.dino","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.1783,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":8.52277,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":122.2,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.5,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.98,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"091622c52c686495","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_base_patch8_224.augreg2_in21k_ft_in1k","title":"ViT Base/8 (224)","source":"timm/vit_base_patch8_224.augreg2_in21k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_base_patch8_224.augreg2_in21k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_base_patch8_224.augreg2_in21k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.08377,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":8.50519,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":123.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":85.06,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.38,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.51,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"198b7bcaf1599bbc","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_betwixt_patch16_reg1_gap_256.sbb_in1k","title":"ViT Betwixt/16 (256, reg1, GAP)","source":"timm/vit_betwixt_patch16_reg1_gap_256.sbb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_betwixt_patch16_reg1_gap_256.sbb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_betwixt_patch16_reg1_gap_256.sbb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.990026,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.28211,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.995575,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.28705,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1008.7,"unit":"qps","digits":1,"baseline":779.078,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.07,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.2,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.68,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"904df70bc4cec3d9","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_betwixt_patch16_reg4_gap_256.sbb2_e200_in12k_ft_in1k","title":"ViT Betwixt/16 (256, reg4, GAP)","source":"timm/vit_betwixt_patch16_reg4_gap_256.sbb2_e200_in12k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_betwixt_patch16_reg4_gap_256.sbb2_e200_in12k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_betwixt_patch16_reg4_gap_256.sbb2_e200_in12k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.986021,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.31397,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.989502,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.31866,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1012.5,"unit":"qps","digits":1,"baseline":760.113,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.86,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.2,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.68,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"07c1a4d6ce4604d9","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_betwixt_patch16_rope_reg4_gap_256.sbb_in1k","title":"ViT Betwixt/16 (256, RoPE, reg4, GAP)","source":"timm/vit_betwixt_patch16_rope_reg4_gap_256.sbb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_betwixt_patch16_rope_reg4_gap_256.sbb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_betwixt_patch16_rope_reg4_gap_256.sbb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.85157,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.85828,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":539.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.4,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.84,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.25,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.73,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"121fb9496d27fda3","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_betwixt_patch16_reg4_gap_384.sbb2_e200_in12k_ft_in1k","title":"ViT Betwixt/16 (384, reg4, GAP)","source":"timm/vit_betwixt_patch16_reg4_gap_384.sbb2_e200_in12k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_betwixt_patch16_reg4_gap_384.sbb2_e200_in12k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_betwixt_patch16_reg4_gap_384.sbb2_e200_in12k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.06618,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.22247,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":326,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":85.2,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.42,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.22,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.69,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"cff57f98c1581b0f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_betwixt_patch32_clip_224.tinyclip_laion400m","title":"ViT Betwixt/32 (224, CLIP)","source":"timm/vit_betwixt_patch32_clip_224.tinyclip_laion400m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_betwixt_patch32_clip_224.tinyclip_laion400m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_betwixt_patch32_clip_224.tinyclip_laion400m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.687594,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.798143,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.689484,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.800415,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1451,"unit":"qps","digits":1,"baseline":1250.81,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.07,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.55,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"96e93c6f081fe1dd","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_giant_patch14_clip_224.laion2b","title":"ViT Giant/14 (224, CLIP)","source":"timm/vit_giant_patch14_clip_224.laion2b","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_giant_patch14_clip_224.laion2b&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_giant_patch14_clip_224.laion2b","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":14.8592,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":15.0139,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":67.3,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":12.04,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":11.51,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"603d08743a4fb642","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_giant_patch14_dinov2.lvd142m","title":"ViT Giant/14 (DINOv2)","source":"timm/vit_giant_patch14_dinov2.lvd142m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_giant_patch14_dinov2.lvd142m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_giant_patch14_dinov2.lvd142m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":202.609,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":209.695,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":4.9,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":13.49,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":12.97,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"e01c5acad2a83fa9","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_giant_patch14_reg1_tipsv2.webli","title":"ViT Giant/14 (reg1, tipsv2)","source":"timm/vit_giant_patch14_reg1_tipsv2.webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_giant_patch14_reg1_tipsv2.webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_giant_patch14_reg1_tipsv2.webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":121.499,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":125.305,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":8.2,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":13.5,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":12.98,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"c96877bb16ad401f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_giant_patch14_reg4_dinov2.lvd142m","title":"ViT Giant/14 (reg4, DINOv2)","source":"timm/vit_giant_patch14_reg4_dinov2.lvd142m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_giant_patch14_reg4_dinov2.lvd142m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_giant_patch14_reg4_dinov2.lvd142m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":201.239,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":205.738,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":5,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":13.48,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":12.96,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"8d0405614a4997d9","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_giant_patch16_lingbot.robbyant","title":"ViT Giant/16 (lingbot)","source":"timm/vit_giant_patch16_lingbot.robbyant","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_giant_patch16_lingbot.robbyant&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_giant_patch16_lingbot.robbyant","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":152.239,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":157.162,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":6.6,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":14.13,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":13.61,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"e74601cf90657c24","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_giant_patch16_sapiens2.fb","title":"ViT Giant/16 (sapiens2)","source":"timm/vit_giant_patch16_sapiens2.fb","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_giant_patch16_sapiens2.fb&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_giant_patch16_sapiens2.fb","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1268.52,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1291.07,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":0.8,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":20.6,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":20.08,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"1c8395dfb0bcc7d1","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_giantopt_patch16_siglip_gap_256.v2_webli","title":"ViT GiantOpt/16 (256, GAP, SigLIP)","source":"timm/vit_giantopt_patch16_siglip_gap_256.v2_webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_giantopt_patch16_siglip_gap_256.v2_webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_giantopt_patch16_siglip_gap_256.v2_webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":12.4313,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":15.5763,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":12.8799,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":16.6326,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":80.4,"unit":"qps","digits":1,"baseline":64.1933,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":13.47,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":12.95,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"afe2ba8702eeffae","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_giantopt_patch16_siglip_256.v2_webli","title":"ViT GiantOpt/16 (256, SigLIP)","source":"timm/vit_giantopt_patch16_siglip_256.v2_webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_giantopt_patch16_siglip_256.v2_webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_giantopt_patch16_siglip_256.v2_webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":12.4474,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":15.6482,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":12.8498,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":16.9126,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":80.3,"unit":"qps","digits":1,"baseline":63.8987,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":13.79,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":13.27,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"fe26d7106d7396a1","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_giantopt_patch16_siglip_gap_384.v2_webli","title":"ViT GiantOpt/16 (384, GAP, SigLIP)","source":"timm/vit_giantopt_patch16_siglip_gap_384.v2_webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_giantopt_patch16_siglip_gap_384.v2_webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_giantopt_patch16_siglip_gap_384.v2_webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":31.9273,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":34.6616,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":31.3,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":13.47,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":12.95,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"170b731e22032b40","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_giantopt_patch16_siglip_384.v2_webli","title":"ViT GiantOpt/16 (384, SigLIP)","source":"timm/vit_giantopt_patch16_siglip_384.v2_webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_giantopt_patch16_siglip_384.v2_webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_giantopt_patch16_siglip_384.v2_webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":32.0119,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":35.0742,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":31.2,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":13.8,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":13.28,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"a965af57b7e89baa","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_gigantic_patch14_clip_224.laion2b","title":"ViT Gigantic/14 (224, CLIP)","source":"timm/vit_gigantic_patch14_clip_224.laion2b","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_gigantic_patch14_clip_224.laion2b&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_gigantic_patch14_clip_224.laion2b","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":24.6699,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":29.2162,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":25.5806,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":31.0085,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":40.5,"unit":"qps","digits":1,"baseline":34.2257,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":21.22,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":20.7,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"16b19f849f704fde","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_gigantic_patch14_clip_quickgelu_224.metaclip_2pt5b","title":"ViT Gigantic/14 (224, QuickGELU, CLIP)","source":"timm/vit_gigantic_patch14_clip_quickgelu_224.metaclip_2pt5b","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_gigantic_patch14_clip_quickgelu_224.metaclip_2pt5b&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_gigantic_patch14_clip_quickgelu_224.metaclip_2pt5b","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":25.6816,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":31.5885,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":26.7736,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":33.4572,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":38.9,"unit":"qps","digits":1,"baseline":31.6555,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":21.44,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":20.92,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"56d5da3965629add","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_gigantic_patch14_clip_378.metaclip2_worldwide","title":"ViT Gigantic/14 (378, CLIP)","source":"timm/vit_gigantic_patch14_clip_378.metaclip2_worldwide","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_gigantic_patch14_clip_378.metaclip2_worldwide&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_gigantic_patch14_clip_378.metaclip2_worldwide","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":68.3087,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":69.9583,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":14.6,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":21.53,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":21.01,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"5a110bd4db07aa30","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_pe_core_gigantic_patch14_448.fb","title":"ViT Gigantic/14 (448, PE, Core)","source":"timm/vit_pe_core_gigantic_patch14_448.fb","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_pe_core_gigantic_patch14_448.fb&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_pe_core_gigantic_patch14_448.fb","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":120.406,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":122.81,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":8.3,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":21.84,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":21.32,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"bbfc45d5820eaab7","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_pe_lang_gigantic_patch14_448.fb","title":"ViT Gigantic/14 (448, PE, Lang)","source":"timm/vit_pe_lang_gigantic_patch14_448.fb","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_pe_lang_gigantic_patch14_448.fb&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_pe_lang_gigantic_patch14_448.fb","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":112.774,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":115.177,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":8.9,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":14.27,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":13.75,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"c13bb1e1ac62105b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_pe_spatial_gigantic_patch14_448.fb","title":"ViT Gigantic/14 (448, PE, Spatial)","source":"timm/vit_pe_spatial_gigantic_patch14_448.fb","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_pe_spatial_gigantic_patch14_448.fb&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_pe_spatial_gigantic_patch14_448.fb","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":118.945,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":120.98,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":8.4,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":21.52,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":21,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"d62de7942306a69d","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_huge_plus_patch16_dinov3.lvd1689m","title":"ViT Huge Plus/16 (DINOv3)","source":"timm/vit_huge_plus_patch16_dinov3.lvd1689m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_huge_plus_patch16_dinov3.lvd1689m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_huge_plus_patch16_dinov3.lvd1689m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":15.4374,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":16.3651,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":64.8,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":10.48,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":9.96,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"af07de46ac1e0566","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_huge_plus_patch16_dinov3_qkvb.lvd1689m","title":"ViT Huge Plus/16 (qkvb, DINOv3)","source":"timm/vit_huge_plus_patch16_dinov3_qkvb.lvd1689m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_huge_plus_patch16_dinov3_qkvb.lvd1689m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_huge_plus_patch16_dinov3_qkvb.lvd1689m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":19.5953,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":21.0339,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":51,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":10.9,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":10.38,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"7d42ebedea712bea","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_huge_patch14_clip_224.dfn5b","title":"ViT Huge/14 (224, CLIP, DFN-5B)","source":"timm/vit_huge_patch14_clip_224.dfn5b","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_huge_patch14_clip_224.dfn5b&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_huge_patch14_clip_224.dfn5b","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.85862,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":10.2331,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":101.4,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":7.72,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":7.2,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"4c07d27a32938d60","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_huge_patch14_clip_224.laion2b_ft_in1k","title":"ViT Huge/14 (224, CLIP, LAION-2B → IN-1k)","source":"timm/vit_huge_patch14_clip_224.laion2b_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_huge_patch14_clip_224.laion2b_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_huge_patch14_clip_224.laion2b_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.59697,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":9.8855,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":104.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":87.01,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.1,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":7.72,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":7.2,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"9b6d465c90a7a967","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_huge_patch14_gap_224.in1k_ijepa","title":"ViT Huge/14 (224, GAP, I-JEPA)","source":"timm/vit_huge_patch14_gap_224.in1k_ijepa","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_huge_patch14_gap_224.in1k_ijepa&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_huge_patch14_gap_224.in1k_ijepa","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.57167,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":8.62873,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":7.79773,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":9.01501,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":132,"unit":"qps","digits":1,"baseline":115.867,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":7.69,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":7.17,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"7ecd3bc01e9c7b47","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_huge_patch14_224.mae","title":"ViT Huge/14 (224, MAE)","source":"timm/vit_huge_patch14_224.mae","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_huge_patch14_224.mae&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_huge_patch14_224.mae","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.77752,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":10.2747,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":102.3,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":7.71,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":7.19,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"a00734baadf192e0","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_huge_patch14_clip_quickgelu_224.dfn5b","title":"ViT Huge/14 (224, QuickGELU, CLIP)","source":"timm/vit_huge_patch14_clip_quickgelu_224.dfn5b","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_huge_patch14_clip_quickgelu_224.dfn5b&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_huge_patch14_clip_quickgelu_224.dfn5b","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.60846,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":10.019,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":104.1,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":7.72,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":7.2,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"9439562be9e6cc3d","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_huge_patch14_clip_336.laion2b_ft_in12k_in1k","title":"ViT Huge/14 (336, CLIP)","source":"timm/vit_huge_patch14_clip_336.laion2b_ft_in12k_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_huge_patch14_clip_336.laion2b_ft_in12k_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_huge_patch14_clip_336.laion2b_ft_in12k_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":23.4438,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":24.4347,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":42.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":88.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.54,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":7.73,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":7.21,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"4d353b8b39535e37","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_huge_patch14_clip_378.dfn5b","title":"ViT Huge/14 (378, CLIP)","source":"timm/vit_huge_patch14_clip_378.dfn5b","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_huge_patch14_clip_378.dfn5b&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_huge_patch14_clip_378.dfn5b","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":33.4814,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":34.072,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":29.9,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":7.72,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":7.2,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"5f36aaceaa93bd1f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_huge_patch14_clip_quickgelu_378.dfn5b","title":"ViT Huge/14 (378, QuickGELU, CLIP)","source":"timm/vit_huge_patch14_clip_quickgelu_378.dfn5b","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_huge_patch14_clip_quickgelu_378.dfn5b&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_huge_patch14_clip_quickgelu_378.dfn5b","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":33.0753,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":33.9064,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":30.2,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":7.72,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":7.2,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"98d7b63d336ff1e8","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_huge_patch16_gap_448.in1k_ijepa","title":"ViT Huge/16 (448, GAP, I-JEPA)","source":"timm/vit_huge_patch16_gap_448.in1k_ijepa","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_huge_patch16_gap_448.in1k_ijepa&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_huge_patch16_gap_448.in1k_ijepa","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":35.5498,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":36.8417,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":28.1,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":7.71,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":7.19,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"f4edf574f5d1a5c3","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_huge_patch16_sapiens2.fb","title":"ViT Huge/16 (sapiens2)","source":"timm/vit_huge_patch16_sapiens2.fb","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_huge_patch16_sapiens2.fb&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_huge_patch16_sapiens2.fb","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":670.845,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":691.117,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1.5,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":11.77,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":11.24,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"b5c19027ddd7b79c","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_intern300m_patch14_448.ogvl_2pt5","title":"ViT Intern-300M/14 (448)","source":"timm/vit_intern300m_patch14_448.ogvl_2pt5","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_intern300m_patch14_448.ogvl_2pt5&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_intern300m_patch14_448.ogvl_2pt5","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":44.2036,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":45.5583,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":22.6,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.95,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.43,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"88b9ebc4a9ca86aa","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch14_clip_224.apple_mclip2_dfndr2b","title":"ViT Large/14 (224, CLIP, Apple, mclip2, dfndr2b)","source":"timm/vit_large_patch14_clip_224.apple_mclip2_dfndr2b","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch14_clip_224.apple_mclip2_dfndr2b&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch14_clip_224.apple_mclip2_dfndr2b","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.55279,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.29028,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":3.6908,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.49841,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":281.3,"unit":"qps","digits":1,"baseline":188.959,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.93,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.4,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"97bfe8a670186198","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch14_clip_224.laion2b_ft_in1k","title":"ViT Large/14 (224, CLIP, LAION-2B → IN-1k)","source":"timm/vit_large_patch14_clip_224.laion2b_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch14_clip_224.laion2b_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch14_clip_224.laion2b_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.67991,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.13098,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":3.76428,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.42413,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":271.6,"unit":"qps","digits":1,"baseline":194.833,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":85.69,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.73,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.4,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"c19080b1119a40b5","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch14_clip_224.openai","title":"ViT Large/14 (224, CLIP, OpenAI)","source":"timm/vit_large_patch14_clip_224.openai","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch14_clip_224.openai&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch14_clip_224.openai","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.69381,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.37517,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":3.75916,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.57703,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":270.6,"unit":"qps","digits":1,"baseline":185.981,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.24,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.72,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"7760345c03de7942","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch14_clip_quickgelu_224.dfn2b","title":"ViT Large/14 (224, QuickGELU, CLIP, DFN-2B)","source":"timm/vit_large_patch14_clip_quickgelu_224.dfn2b","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch14_clip_quickgelu_224.dfn2b&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch14_clip_quickgelu_224.dfn2b","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.52703,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.17681,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":3.59961,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.37061,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":283.4,"unit":"qps","digits":1,"baseline":193.106,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.93,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.4,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"b96742539bf57fbc","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch14_clip_quickgelu_224.openai","title":"ViT Large/14 (224, QuickGELU, CLIP, OpenAI)","source":"timm/vit_large_patch14_clip_quickgelu_224.openai","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch14_clip_quickgelu_224.openai&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch14_clip_quickgelu_224.openai","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.5075,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.24972,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":3.60132,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.427,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":285,"unit":"qps","digits":1,"baseline":190.429,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.24,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.72,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"790d609a34cf2903","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch14_clip_336.laion2b_ft_in1k","title":"ViT Large/14 (336, CLIP, LAION-2B → IN-1k)","source":"timm/vit_large_patch14_clip_336.laion2b_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch14_clip_336.laion2b_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch14_clip_336.laion2b_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":14.1913,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":14.8512,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":70.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":86.1,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.81,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.99,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.47,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"068cc22b58c96037","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch14_clip_336.openai","title":"ViT Large/14 (336, CLIP, OpenAI)","source":"timm/vit_large_patch14_clip_336.openai","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch14_clip_336.openai&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch14_clip_336.openai","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":14.1901,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":14.8813,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":70.5,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.28,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.76,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"fdfbe3644d1fff03","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_pe_core_large_patch14_336.fb","title":"ViT Large/14 (336, PE, Core)","source":"timm/vit_pe_core_large_patch14_336.fb","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_pe_core_large_patch14_336.fb&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_pe_core_large_patch14_336.fb","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":15.7554,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":16.403,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":63.5,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4.15,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.63,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"47ec50f995d5a31e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch14_clip_quickgelu_336.openai","title":"ViT Large/14 (336, QuickGELU, CLIP)","source":"timm/vit_large_patch14_clip_quickgelu_336.openai","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch14_clip_quickgelu_336.openai&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch14_clip_quickgelu_336.openai","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":14.2102,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":14.8724,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":70.4,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.28,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.76,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"3581686af9aa9ff7","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_pe_lang_large_patch14_448.fb","title":"ViT Large/14 (448, PE, Lang)","source":"timm/vit_pe_lang_large_patch14_448.fb","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_pe_lang_large_patch14_448.fb&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_pe_lang_large_patch14_448.fb","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":44.2629,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":45.1868,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":22.6,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.85,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.33,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"5addba01ce98df25","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_pe_spatial_large_patch14_448.fb","title":"ViT Large/14 (448, PE, Spatial)","source":"timm/vit_pe_spatial_large_patch14_448.fb","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_pe_spatial_large_patch14_448.fb&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_pe_spatial_large_patch14_448.fb","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":47.0001,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":48.0822,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":21.3,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.48,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"4b0f4d2fda48bebb","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch14_dinov2.lvd142m","title":"ViT Large/14 (DINOv2)","source":"timm/vit_large_patch14_dinov2.lvd142m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch14_dinov2.lvd142m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch14_dinov2.lvd142m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":69.2638,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":70.6482,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":14.4,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.47,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"1dd938022e7d859c","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch14_reg1_tipsv2.webli","title":"ViT Large/14 (reg1, tipsv2)","source":"timm/vit_large_patch14_reg1_tipsv2.webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch14_reg1_tipsv2.webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch14_reg1_tipsv2.webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":44.2405,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":45.7432,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":22.6,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.47,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"f950859fe8b2462e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch14_reg4_dinov2.lvd142m","title":"ViT Large/14 (reg4, DINOv2)","source":"timm/vit_large_patch14_reg4_dinov2.lvd142m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch14_reg4_dinov2.lvd142m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch14_reg4_dinov2.lvd142m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":74.0487,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":75.5604,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":13.5,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.48,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"428f4e9ee2594998","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch16_224.mae","title":"ViT Large/16 (224, MAE)","source":"timm/vit_large_patch16_224.mae","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch16_224.mae&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch16_224.mae","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.12183,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.41502,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":3.18689,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.6394,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":320.2,"unit":"qps","digits":1,"baseline":226.418,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.4,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"4b62b61686e71038","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch16_rope_ape_224.naver_in1k","title":"ViT Large/16 (224, RoPE, ape)","source":"timm/vit_large_patch16_rope_ape_224.naver_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch16_rope_ape_224.naver_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch16_rope_ape_224.naver_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.74127,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.81128,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":210.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.56,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.48,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"9506442f8bd3ef31","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch16_rope_224.naver_in1k","title":"ViT Large/16 (224, RoPE)","source":"timm/vit_large_patch16_rope_224.naver_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch16_rope_224.naver_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch16_rope_224.naver_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.73985,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.84393,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":210.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.47,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.97,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.47,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"b2843e4fcb135ed0","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch16_224.augreg_in21k_ft_in1k","title":"ViT Large/16 (224)","source":"timm/vit_large_patch16_224.augreg_in21k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch16_224.augreg_in21k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch16_224.augreg_in21k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.14753,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.4126,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":3.22803,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.57947,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":317.5,"unit":"qps","digits":1,"baseline":226.531,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.67,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.42,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.93,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.41,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"019b5f9372ef17c7","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch16_siglip_gap_256.v2_webli","title":"ViT Large/16 (256, GAP, SigLIP)","source":"timm/vit_large_patch16_siglip_gap_256.v2_webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch16_siglip_gap_256.v2_webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch16_siglip_gap_256.v2_webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.39343,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.85572,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":3.48975,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.06293,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":294.5,"unit":"qps","digits":1,"baseline":205.874,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.92,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.4,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"797efd2eadcc2fb8","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch16_siglip_256.v2_webli","title":"ViT Large/16 (256, SigLIP)","source":"timm/vit_large_patch16_siglip_256.v2_webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch16_siglip_256.v2_webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch16_siglip_256.v2_webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.49474,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.9483,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":3.59357,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":5.21704,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":286,"unit":"qps","digits":1,"baseline":202.024,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4.06,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.54,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"2e3b59da92df8d5b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch16_siglip_gap_384.v2_webli","title":"ViT Large/16 (384, GAP, SigLIP)","source":"timm/vit_large_patch16_siglip_gap_384.v2_webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch16_siglip_gap_384.v2_webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch16_siglip_gap_384.v2_webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":12.2136,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":12.5574,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":81.9,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.99,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.47,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"0d71a90fb7cabc82","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch16_siglip_384.v2_webli","title":"ViT Large/16 (384, SigLIP)","source":"timm/vit_large_patch16_siglip_384.v2_webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch16_siglip_384.v2_webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch16_siglip_384.v2_webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":12.1525,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":12.3889,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":82.3,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4.13,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.61,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"c2648da7a99925f2","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch16_384.augreg_in21k_ft_in1k","title":"ViT Large/16 (384)","source":"timm/vit_large_patch16_384.augreg_in21k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch16_384.augreg_in21k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch16_384.augreg_in21k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":14.0401,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":14.5129,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":71.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":86.27,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.05,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.48,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"12c5a8f570571e8e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch16_siglip_gap_512.v2_webli","title":"ViT Large/16 (512, GAP, SigLIP)","source":"timm/vit_large_patch16_siglip_gap_512.v2_webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch16_siglip_gap_512.v2_webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch16_siglip_gap_512.v2_webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":38.7882,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":40.0533,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":25.8,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.47,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"1cf3a11a80228c61","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch16_siglip_512.v2_webli","title":"ViT Large/16 (512, SigLIP)","source":"timm/vit_large_patch16_siglip_512.v2_webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch16_siglip_512.v2_webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch16_siglip_512.v2_webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":39.0456,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":40.1296,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":25.6,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4.14,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.61,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"3383bed2dc6d0a1c","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch16_dinov3.lvd1689m","title":"ViT Large/16 (DINOv3)","source":"timm/vit_large_patch16_dinov3.lvd1689m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch16_dinov3.lvd1689m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch16_dinov3.lvd1689m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.69709,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.75879,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":270.4,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.91,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.39,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"6a234456f4eca9e7","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch16_lingbot.robbyant","title":"ViT Large/16 (lingbot)","source":"timm/vit_large_patch16_lingbot.robbyant","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch16_lingbot.robbyant&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch16_lingbot.robbyant","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":51.6757,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":52.7513,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":19.4,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4.17,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.65,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"33d4abc13a27ee9e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch16_dinov3_qkvb.lvd1689m","title":"ViT Large/16 (qkvb, DINOv3)","source":"timm/vit_large_patch16_dinov3_qkvb.lvd1689m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch16_dinov3_qkvb.lvd1689m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch16_dinov3_qkvb.lvd1689m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.88772,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.05005,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":169.8,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4.11,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.59,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"8d73e56c27cea554","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch16_sapiens2.fb","title":"ViT Large/16 (sapiens2)","source":"timm/vit_large_patch16_sapiens2.fb","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch16_sapiens2.fb&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch16_sapiens2.fb","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":414.669,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":425.895,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2.4,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":5.98,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":5.46,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"03748048cd38b9cd","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch32_224.orig_in21k","title":"ViT Large/32 (224)","source":"timm/vit_large_patch32_224.orig_in21k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch32_224.orig_in21k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch32_224.orig_in21k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.11495,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.0328,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":2.16034,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.05951,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":472.5,"unit":"qps","digits":1,"baseline":329.542,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.94,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.42,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"42d5b7e2e8218497","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_large_patch32_384.orig_in21k_ft_in1k","title":"ViT Large/32 (384)","source":"timm/vit_large_patch32_384.orig_in21k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_large_patch32_384.orig_in21k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_large_patch32_384.orig_in21k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.02711,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.78182,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":3.09204,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.85077,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":330.2,"unit":"qps","digits":1,"baseline":264.315,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.91,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.78,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.96,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.44,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"9d339afea6f2abf0","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_little_patch16_reg1_gap_256.sbb_in12k_ft_in1k","title":"ViT Little/16 (256, reg1, GAP)","source":"timm/vit_little_patch16_reg1_gap_256.sbb_in12k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_little_patch16_reg1_gap_256.sbb_in12k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_little_patch16_reg1_gap_256.sbb_in12k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.559363,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.7001,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.561401,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.704102,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1782.7,"unit":"qps","digits":1,"baseline":1424.84,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.96,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.63,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.78,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.25,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"ea5d51da6b3bac72","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_little_patch16_reg4_gap_256.sbb_in1k","title":"ViT Little/16 (256, reg4, GAP)","source":"timm/vit_little_patch16_reg4_gap_256.sbb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_little_patch16_reg4_gap_256.sbb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_little_patch16_reg4_gap_256.sbb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.556871,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.712329,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.557983,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.714966,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1791.3,"unit":"qps","digits":1,"baseline":1401.11,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.09,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.46,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.78,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.25,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"5c2d83833536b295","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_mediumd_patch16_reg4_gap_256.sbb2_e200_in12k_ft_in1k","title":"ViT Medium-D/16 (256, reg4, GAP)","source":"timm/vit_mediumd_patch16_reg4_gap_256.sbb2_e200_in12k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_mediumd_patch16_reg4_gap_256.sbb2_e200_in12k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_mediumd_patch16_reg4_gap_256.sbb2_e200_in12k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.23208,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.48061,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.23834,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.5202,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":810.6,"unit":"qps","digits":1,"baseline":674.737,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":85.93,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.79,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.24,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.72,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"92dd8fc4d9a19d37","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_mediumd_patch16_rope_reg1_gap_256.sbb_in1k","title":"ViT Medium-D/16 (256, RoPE, reg1, GAP)","source":"timm/vit_mediumd_patch16_rope_reg1_gap_256.sbb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_mediumd_patch16_rope_reg1_gap_256.sbb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_mediumd_patch16_rope_reg1_gap_256.sbb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.20427,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.27606,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":453.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.68,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.53,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.25,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.73,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"9b1238012571daa2","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_mediumd_patch16_reg4_gap_384.sbb2_e200_in12k_ft_in1k","title":"ViT Medium-D/16 (384, reg4, GAP)","source":"timm/vit_mediumd_patch16_reg4_gap_384.sbb2_e200_in12k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_mediumd_patch16_reg4_gap_384.sbb2_e200_in12k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_mediumd_patch16_reg4_gap_384.sbb2_e200_in12k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.76959,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.90808,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":265.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":87.05,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.13,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.26,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.74,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"a1333ab956414763","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_medium_patch16_clip_224.tinyclip_yfcc15m","title":"ViT Medium/16 (224, CLIP)","source":"timm/vit_medium_patch16_clip_224.tinyclip_yfcc15m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_medium_patch16_clip_224.tinyclip_yfcc15m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_medium_patch16_clip_224.tinyclip_yfcc15m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.572365,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.798748,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.57373,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.805115,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1743,"unit":"qps","digits":1,"baseline":1249.73,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.87,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.35,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"c14e9a98ae6a3d87","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_relpos_medium_patch16_cls_224.sw_in1k","title":"ViT Medium/16 (224, relpos, cls)","source":"timm/vit_relpos_medium_patch16_cls_224.sw_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_relpos_medium_patch16_cls_224.sw_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_relpos_medium_patch16_cls_224.sw_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.841846,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.05049,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.846344,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.05383,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1185.6,"unit":"qps","digits":1,"baseline":950.697,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.04,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.83,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.96,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.44,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"173916899d6a3f9e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_relpos_medium_patch16_rpn_224.sw_in1k","title":"ViT Medium/16 (224, relpos, rpn)","source":"timm/vit_relpos_medium_patch16_rpn_224.sw_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_relpos_medium_patch16_rpn_224.sw_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_relpos_medium_patch16_rpn_224.sw_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.730318,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.928643,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.733704,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.930664,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1366.3,"unit":"qps","digits":1,"baseline":1075.05,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.66,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.67,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.96,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.44,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"02c204c3fa59a190","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_relpos_medium_patch16_224.sw_in1k","title":"ViT Medium/16 (224, relpos)","source":"timm/vit_relpos_medium_patch16_224.sw_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_relpos_medium_patch16_224.sw_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_relpos_medium_patch16_224.sw_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.726549,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.935147,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.729248,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.938354,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1373.5,"unit":"qps","digits":1,"baseline":1067.58,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.31,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.87,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.96,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.44,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"8d5410ee3e506e9b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_srelpos_medium_patch16_224.sw_in1k","title":"ViT Medium/16 (224, srelpos)","source":"timm/vit_srelpos_medium_patch16_224.sw_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_srelpos_medium_patch16_224.sw_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_srelpos_medium_patch16_224.sw_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.571747,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.855339,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.573822,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.858032,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1743.6,"unit":"qps","digits":1,"baseline":1166.73,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.74,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.71,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.96,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.44,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"ba3e0124820a557f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_medium_patch16_gap_240.sw_in12k","title":"ViT Medium/16 (240, GAP)","source":"timm/vit_medium_patch16_gap_240.sw_in12k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_medium_patch16_gap_240.sw_in12k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_medium_patch16_gap_240.sw_in12k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.597292,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.820103,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.598999,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.827759,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1670.4,"unit":"qps","digits":1,"baseline":1217.3,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.95,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.43,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"ea2d4d89f173ba98","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_medium_patch16_gap_256.sw_in12k_ft_in1k","title":"ViT Medium/16 (256, GAP)","source":"timm/vit_medium_patch16_gap_256.sw_in12k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_medium_patch16_gap_256.sw_in12k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_medium_patch16_gap_256.sw_in12k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.61695,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.848678,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.618774,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.851135,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1617.3,"unit":"qps","digits":1,"baseline":1175.86,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.07,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.71,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.96,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.44,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"589ef09b22117d40","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_medium_patch16_reg1_gap_256.sbb_in1k","title":"ViT Medium/16 (256, reg1, GAP)","source":"timm/vit_medium_patch16_reg1_gap_256.sbb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_medium_patch16_reg1_gap_256.sbb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_medium_patch16_reg1_gap_256.sbb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.690403,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.91948,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.692505,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.921143,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1445.6,"unit":"qps","digits":1,"baseline":1085.91,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.15,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.91,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.96,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.44,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"bb47f5acd6c9ccae","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_medium_patch16_rope_reg1_gap_256.sbb_in1k","title":"ViT Medium/16 (256, RoPE, reg1, GAP)","source":"timm/vit_medium_patch16_rope_reg1_gap_256.sbb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_medium_patch16_rope_reg1_gap_256.sbb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_medium_patch16_rope_reg1_gap_256.sbb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.54538,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.55292,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":646.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.37,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.53,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.99,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.47,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"4f5ffa419ded92b2","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_medium_patch16_gap_384.sw_in12k_ft_in1k","title":"ViT Medium/16 (384, GAP)","source":"timm/vit_medium_patch16_gap_384.sw_in12k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_medium_patch16_gap_384.sw_in12k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_medium_patch16_gap_384.sw_in12k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.20092,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.31073,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":454,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.67,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.83,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.97,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.45,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e16dfb2e5f3b9b09","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_medium_patch32_clip_224.tinyclip_laion400m","title":"ViT Medium/32 (224, CLIP)","source":"timm/vit_medium_patch32_clip_224.tinyclip_laion400m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_medium_patch32_clip_224.tinyclip_laion400m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_medium_patch32_clip_224.tinyclip_laion400m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.621908,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.624329,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1603.9,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.88,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.36,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"56dd7c8054377ebb","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_pwee_patch16_reg1_gap_256.sbb_in1k","title":"ViT PWee/16 (256, reg1, GAP)","source":"timm/vit_pwee_patch16_reg1_gap_256.sbb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_pwee_patch16_reg1_gap_256.sbb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_pwee_patch16_reg1_gap_256.sbb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.493548,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.602306,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.494507,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.605713,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":2020.6,"unit":"qps","digits":1,"baseline":1655.5,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.84,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.4,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.69,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.17,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"ced3853dda26c80b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_small_plus_patch16_dinov3.lvd1689m","title":"ViT Small Plus/16 (DINOv3)","source":"timm/vit_small_plus_patch16_dinov3.lvd1689m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_small_plus_patch16_dinov3.lvd1689m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_small_plus_patch16_dinov3.lvd1689m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.918817,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.922302,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1086.4,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.86,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.34,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"189b960dfd274f6a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_small_plus_patch16_dinov3_qkvb.lvd1689m","title":"ViT Small Plus/16 (qkvb, DINOv3)","source":"timm/vit_small_plus_patch16_dinov3_qkvb.lvd1689m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_small_plus_patch16_dinov3_qkvb.lvd1689m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_small_plus_patch16_dinov3_qkvb.lvd1689m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.06157,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.06586,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":940.4,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.88,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.36,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"d3435ea2d99448ec","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_small_patch14_dinov2.lvd142m","title":"ViT Small/14 (DINOv2)","source":"timm/vit_small_patch14_dinov2.lvd142m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_small_patch14_dinov2.lvd142m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_small_patch14_dinov2.lvd142m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":11.5424,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":11.9883,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":86.6,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.78,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.25,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"5d0948ad3057446c","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_small_patch14_reg4_dinov2.lvd142m","title":"ViT Small/14 (reg4, DINOv2)","source":"timm/vit_small_patch14_reg4_dinov2.lvd142m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_small_patch14_reg4_dinov2.lvd142m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_small_patch14_reg4_dinov2.lvd142m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":11.673,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":12.054,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":85.7,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.78,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.25,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"7d7efc4aa2c4c0a0","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_small_patch16_224.dino","title":"ViT Small/16 (224, DINO)","source":"timm/vit_small_patch16_224.dino","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_small_patch16_224.dino&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_small_patch16_224.dino","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.466691,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.517897,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.468018,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.520508,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":2136,"unit":"qps","digits":1,"baseline":1925.07,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.54,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":19.3,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image"],"build":"2026102013360c6f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_relpos_small_patch16_224.sw_in1k","title":"ViT Small/16 (224, relpos)","source":"timm/vit_relpos_small_patch16_224.sw_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_relpos_small_patch16_224.sw_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_relpos_small_patch16_224.sw_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.502196,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.674224,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.503296,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.677979,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1985,"unit":"qps","digits":1,"baseline":1480.13,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.11,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.7,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.77,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.25,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"f8fd5a541c8bc607","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_small_patch16_rope_ape_224.naver_in1k","title":"ViT Small/16 (224, RoPE, ape)","source":"timm/vit_small_patch16_rope_ape_224.naver_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_small_patch16_rope_ape_224.naver_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_small_patch16_rope_ape_224.naver_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.755322,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.757751,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1321.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":79.73,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.92,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.78,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.25,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"b0e301e0f2dbd6da","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_small_patch16_rope_224.naver_in1k","title":"ViT Small/16 (224, RoPE)","source":"timm/vit_small_patch16_rope_224.naver_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_small_patch16_rope_224.naver_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_small_patch16_rope_224.naver_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.775447,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.778259,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1286.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.29,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.44,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.77,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.25,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"de19a369acbbc832","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_srelpos_small_patch16_224.sw_in1k","title":"ViT Small/16 (224, srelpos)","source":"timm/vit_srelpos_small_patch16_224.sw_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_srelpos_small_patch16_224.sw_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_srelpos_small_patch16_224.sw_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.468979,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.571599,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.46991,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.572876,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":2126,"unit":"qps","digits":1,"baseline":1744.75,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.61,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.32,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.77,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.25,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"7ff88bb683dd13e8","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_small_patch16_224.augreg_in1k","title":"ViT Small/16 (224)","source":"timm/vit_small_patch16_224.augreg_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_small_patch16_224.augreg_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_small_patch16_224.augreg_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.468184,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.522973,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.469055,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.526794,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":2128.7,"unit":"qps","digits":1,"baseline":1906.49,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":77.95,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.85,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.77,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.25,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"05e55f60af7b2e57","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_pe_core_small_patch16_384.fb","title":"ViT Small/16 (384, PE, Core)","source":"timm/vit_pe_core_small_patch16_384.fb","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_pe_core_small_patch16_384.fb&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_pe_core_small_patch16_384.fb","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.32748,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.39087,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":429.4,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.8,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.27,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"0540dba307034aac","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_small_patch16_384.augreg_in1k","title":"ViT Small/16 (384)","source":"timm/vit_small_patch16_384.augreg_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_small_patch16_384.augreg_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_small_patch16_384.augreg_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.07436,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.16571,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":481.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.49,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.33,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.78,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.26,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"2ef978470e070d20","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_pe_spatial_small_patch16_512.fb","title":"ViT Small/16 (512, PE, Spatial)","source":"timm/vit_pe_spatial_small_patch16_512.fb","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_pe_spatial_small_patch16_512.fb&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_pe_spatial_small_patch16_512.fb","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.79288,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.10779,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":172.6,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.78,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.25,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"a966d34538a53773","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_small_patch16_dinov3.lvd1689m","title":"ViT Small/16 (DINOv3)","source":"timm/vit_small_patch16_dinov3.lvd1689m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_small_patch16_dinov3.lvd1689m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_small_patch16_dinov3.lvd1689m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.514321,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.516785,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1938.4,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.76,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.24,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"835af2f7b7578d29","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_small_patch16_lingbot.robbyant","title":"ViT Small/16 (lingbot)","source":"timm/vit_small_patch16_lingbot.robbyant","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_small_patch16_lingbot.robbyant&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_small_patch16_lingbot.robbyant","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":6.21013,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":6.39844,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":161,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.78,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.26,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"2122d84313bcd97f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_small_patch16_dinov3_qkvb.eupe_lvd1689m","title":"ViT Small/16 (qkvb, DINOv3)","source":"timm/vit_small_patch16_dinov3_qkvb.eupe_lvd1689m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_small_patch16_dinov3_qkvb.eupe_lvd1689m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_small_patch16_dinov3_qkvb.eupe_lvd1689m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.649684,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.65155,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1535,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.78,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.26,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"9a3938354b003e95","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_small_patch32_224.augreg_in21k_ft_in1k","title":"ViT Small/32 (224)","source":"timm/vit_small_patch32_224.augreg_in21k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_small_patch32_224.augreg_in21k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_small_patch32_224.augreg_in21k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.434864,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.435425,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2291.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":73.96,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.18,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.78,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.26,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"24ea58eecf499f38","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_small_patch32_384.augreg_in21k_ft_in1k","title":"ViT Small/32 (384)","source":"timm/vit_small_patch32_384.augreg_in21k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_small_patch32_384.augreg_in21k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_small_patch32_384.augreg_in21k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.509239,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.51062,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1958.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.13,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.72,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.78,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.26,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"a940e4c445471d3c","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_small_patch8_224.dino","title":"ViT Small/8 (224, DINO)","source":"timm/vit_small_patch8_224.dino","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_small_patch8_224.dino&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_small_patch8_224.dino","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.45112,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.61346,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":289.6,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.77,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.25,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"6f8741451d45a950","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_so150m_patch16_reg4_gap_256.sbb_e250_in12k_ft_in1k","title":"ViT SO150M/16 (256, reg4, GAP)","source":"timm/vit_so150m_patch16_reg4_gap_256.sbb_e250_in12k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_so150m_patch16_reg4_gap_256.sbb_e250_in12k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_so150m_patch16_reg4_gap_256.sbb_e250_in12k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.90544,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.53065,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.91968,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.65771,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":524.4,"unit":"qps","digits":1,"baseline":394.911,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.81,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.38,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.03,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.5,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"4607d23bf1917ece","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_so150m_patch16_reg4_gap_384.sbb_e250_in12k_ft_in1k","title":"ViT SO150M/16 (384, reg4, GAP)","source":"timm/vit_so150m_patch16_reg4_gap_384.sbb_e250_in12k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_so150m_patch16_reg4_gap_384.sbb_e250_in12k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_so150m_patch16_reg4_gap_384.sbb_e250_in12k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.30698,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":7.60364,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":136.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":86.31,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.07,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.55,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"3cbe649cd1436600","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_so150m2_patch16_reg1_gap_256.sbb_e200_in12k_ft_in1k","title":"ViT SO150M2/16 (256, reg1, GAP)","source":"timm/vit_so150m2_patch16_reg1_gap_256.sbb_e200_in12k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_so150m2_patch16_reg1_gap_256.sbb_e200_in12k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_so150m2_patch16_reg1_gap_256.sbb_e200_in12k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.05612,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.76117,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":2.06104,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.85663,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":486,"unit":"qps","digits":1,"baseline":361.968,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.79,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.82,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.05,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.53,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"34ffb23b3578e7dc","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_so150m2_patch16_reg1_gap_384.sbb_e200_in12k_ft_in1k","title":"ViT SO150M2/16 (384, reg1, GAP)","source":"timm/vit_so150m2_patch16_reg1_gap_384.sbb_e200_in12k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_so150m2_patch16_reg1_gap_384.sbb_e200_in12k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_so150m2_patch16_reg1_gap_384.sbb_e200_in12k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.28491,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":9.6781,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":107.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.74,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.59,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.09,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.57,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e0f036a25c690187","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_so150m2_patch16_reg1_gap_448.sbb_e200_in12k_ft_in1k","title":"ViT SO150M2/16 (448, reg1, GAP)","source":"timm/vit_so150m2_patch16_reg1_gap_448.sbb_e200_in12k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_so150m2_patch16_reg1_gap_448.sbb_e200_in12k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_so150m2_patch16_reg1_gap_448.sbb_e200_in12k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":15.0731,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":16.0942,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":66.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.67,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.67,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.09,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.57,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"efc5750b43db7396","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_so400m_patch14_siglip_gap_224.pali2_3b_pt","title":"ViT SO400M/14 (224, GAP, SigLIP)","source":"timm/vit_so400m_patch14_siglip_gap_224.pali2_3b_pt","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_so400m_patch14_siglip_gap_224.pali2_3b_pt&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_so400m_patch14_siglip_gap_224.pali2_3b_pt","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.64645,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":5.84296,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":177.1,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.83,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.31,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"8731ae61e1c731c3","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_so400m_patch14_siglip_224.v2_webli","title":"ViT SO400M/14 (224, SigLIP)","source":"timm/vit_so400m_patch14_siglip_224.v2_webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_so400m_patch14_siglip_224.v2_webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_so400m_patch14_siglip_224.v2_webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.83077,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":6.40449,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":5.97449,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":6.90326,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":171.5,"unit":"qps","digits":1,"baseline":156.099,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":5.41,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":4.88,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"67ef98eafb5aa5fa","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_so400m_patch14_siglip_gap_378.webli_ft_in1k","title":"ViT SO400M/14 (378, GAP, SigLIP, WebLI → IN-1k)","source":"timm/vit_so400m_patch14_siglip_gap_378.webli_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_so400m_patch14_siglip_gap_378.webli_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_so400m_patch14_siglip_gap_378.webli_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":25.6052,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":25.8861,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":39.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":88.76,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.72,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":5.25,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":4.73,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"2ae6dacf5763e58e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_so400m_patch14_siglip_gap_378.v2_webli","title":"ViT SO400M/14 (378, GAP, SigLIP, WebLI v2)","source":"timm/vit_so400m_patch14_siglip_gap_378.v2_webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_so400m_patch14_siglip_gap_378.v2_webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_so400m_patch14_siglip_gap_378.v2_webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":25.2738,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":26.0991,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":39.6,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":5.24,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":4.72,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"7160c0457fa611f8","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_so400m_patch14_siglip_378.webli_ft_in1k","title":"ViT SO400M/14 (378, SigLIP, WebLI → IN-1k)","source":"timm/vit_so400m_patch14_siglip_378.webli_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_so400m_patch14_siglip_378.webli_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_so400m_patch14_siglip_378.webli_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":25.4401,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":26.0114,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":39.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":89.17,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":98.79,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":5.43,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":4.91,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"f89ac4d24eb0d9f4","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_so400m_patch14_siglip_378.v2_webli","title":"ViT SO400M/14 (378, SigLIP, WebLI v2)","source":"timm/vit_so400m_patch14_siglip_378.v2_webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_so400m_patch14_siglip_378.v2_webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_so400m_patch14_siglip_378.v2_webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":25.1631,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":25.823,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":39.7,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":5.41,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":4.89,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"3e12ba10266d8a75","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_so400m_patch14_siglip_gap_384.webli","title":"ViT SO400M/14 (384, GAP, SigLIP)","source":"timm/vit_so400m_patch14_siglip_gap_384.webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_so400m_patch14_siglip_gap_384.webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_so400m_patch14_siglip_gap_384.webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":24.9732,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":25.4003,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":40,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":5.24,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":4.72,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"492223a45738d080","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_so400m_patch14_siglip_384.webli","title":"ViT SO400M/14 (384, SigLIP)","source":"timm/vit_so400m_patch14_siglip_384.webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_so400m_patch14_siglip_384.webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_so400m_patch14_siglip_384.webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":25.7472,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":26.1422,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":38.8,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":5.42,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":4.9,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"a45c737cf9396fb8","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_so400m_patch14_siglip_gap_448.pali2_3b_docci","title":"ViT SO400M/14 (448, GAP, SigLIP)","source":"timm/vit_so400m_patch14_siglip_gap_448.pali2_3b_docci","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_so400m_patch14_siglip_gap_448.pali2_3b_docci&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_so400m_patch14_siglip_gap_448.pali2_3b_docci","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":50.1946,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":51.1856,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":19.9,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.83,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.31,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"0905da367f5851cd","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_so400m_patch14_siglip_gap_896.pali2_3b_pt","title":"ViT SO400M/14 (896, GAP, SigLIP)","source":"timm/vit_so400m_patch14_siglip_gap_896.pali2_3b_pt","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_so400m_patch14_siglip_gap_896.pali2_3b_pt&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_so400m_patch14_siglip_gap_896.pali2_3b_pt","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":603.608,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":633.695,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1.7,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":3.86,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.34,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"6891945ac991fd2c","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_so400m_patch14_reg1_tipsv2.webli","title":"ViT SO400M/14 (reg1, tipsv2)","source":"timm/vit_so400m_patch14_reg1_tipsv2.webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_so400m_patch14_reg1_tipsv2.webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_so400m_patch14_reg1_tipsv2.webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":52.1953,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":53.2929,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":19.2,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":5.25,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":4.73,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"594059cc551b2d53","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_so400m_patch16_siglip_gap_256.v2_webli","title":"ViT SO400M/16 (256, GAP, SigLIP)","source":"timm/vit_so400m_patch16_siglip_gap_256.v2_webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_so400m_patch16_siglip_gap_256.v2_webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_so400m_patch16_siglip_gap_256.v2_webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.60116,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":6.59014,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":5.85364,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":7.03815,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":178.5,"unit":"qps","digits":1,"baseline":151.7,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":5.23,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":4.71,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"59da6aa4939b8980","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_so400m_patch16_siglip_256.v2_webli","title":"ViT SO400M/16 (256, SigLIP)","source":"timm/vit_so400m_patch16_siglip_256.v2_webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_so400m_patch16_siglip_256.v2_webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_so400m_patch16_siglip_256.v2_webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.72594,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":6.69886,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":5.87231,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":7.08337,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":174.6,"unit":"qps","digits":1,"baseline":149.244,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":5.41,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":4.89,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"29a355c8144a8d78","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_so400m_patch16_siglip_gap_384.v2_webli","title":"ViT SO400M/16 (384, GAP, SigLIP)","source":"timm/vit_so400m_patch16_siglip_gap_384.v2_webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_so400m_patch16_siglip_gap_384.v2_webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_so400m_patch16_siglip_gap_384.v2_webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":15.2883,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":15.7432,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":65.4,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":5.24,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":4.72,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"c328a7585a96f8b5","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_so400m_patch16_siglip_384.v2_webli","title":"ViT SO400M/16 (384, SigLIP)","source":"timm/vit_so400m_patch16_siglip_384.v2_webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_so400m_patch16_siglip_384.v2_webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_so400m_patch16_siglip_384.v2_webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":14.8994,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":15.5956,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":67.1,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":5.41,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":4.89,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"dd0be817650ab065","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_so400m_patch16_siglip_gap_512.v2_webli","title":"ViT SO400M/16 (512, GAP, SigLIP)","source":"timm/vit_so400m_patch16_siglip_gap_512.v2_webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_so400m_patch16_siglip_gap_512.v2_webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_so400m_patch16_siglip_gap_512.v2_webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":49.7331,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":50.6353,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":20.1,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":5.25,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":4.72,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"3c2037c65ac9eb5f","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_so400m_patch16_siglip_512.v2_webli","title":"ViT SO400M/16 (512, SigLIP)","source":"timm/vit_so400m_patch16_siglip_512.v2_webli","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_so400m_patch16_siglip_512.v2_webli&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_so400m_patch16_siglip_512.v2_webli","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":49.823,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":50.7805,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":20.1,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":5.42,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":4.9,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"45f1ed07fbb22203","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_tiny_patch16_224.augreg_in21k_ft_in1k","title":"ViT Tiny/16 (224)","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_tiny_patch16_224.augreg_in21k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_tiny_patch16_224.augreg_in21k_ft_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":344.421,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":344.849,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":2890.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":69.71,"unit":"%","headline":true},{"description":"Top-5","value":89.73,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_pe_core_tiny_patch16_384.fb","title":"ViT Tiny/16 (384, PE, Core)","source":"timm/vit_pe_core_tiny_patch16_384.fb","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_pe_core_tiny_patch16_384.fb&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_pe_core_tiny_patch16_384.fb","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.21258,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.21716,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":823.6,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":73.7,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image"],"build":"07012f1e62287bd0","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_tiny_patch16_384.augreg_in21k_ft_in1k","title":"ViT Tiny/16 (384)","source":"timm/vit_tiny_patch16_384.augreg_in21k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_tiny_patch16_384.augreg_in21k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_tiny_patch16_384.augreg_in21k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.05386,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.05908,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":947.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":74.21,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.55,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":68.9,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"1f235227f0031b7a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_pe_spatial_tiny_patch16_512.fb","title":"ViT Tiny/16 (512, PE, Spatial)","source":"timm/vit_pe_spatial_tiny_patch16_512.fb","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_pe_spatial_tiny_patch16_512.fb&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_pe_spatial_tiny_patch16_512.fb","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.03118,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.18262,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":329.7,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":68.3,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image"],"build":"e38c87c7b2bf14a3","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_tiny_patch16_dinov3_qkvb.eupe_lvd1689m","title":"ViT Tiny/16 (qkvb, DINOv3)","source":"timm/vit_tiny_patch16_dinov3_qkvb.eupe_lvd1689m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_tiny_patch16_dinov3_qkvb.eupe_lvd1689m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_tiny_patch16_dinov3_qkvb.eupe_lvd1689m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.37935,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.380249,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2625.9,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.59,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":67.7,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image"],"build":"6402c07b7f83e5f4","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_tiny_r_s16_p8_384.augreg_in21k_ft_in1k","title":"ViT Tiny/8 (384, r, s16)","source":"timm/vit_tiny_r_s16_p8_384.augreg_in21k_ft_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_tiny_r_s16_p8_384.augreg_in21k_ft_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_tiny_r_s16_p8_384.augreg_in21k_ft_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.2.1.2, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.429128,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.431335,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":2322.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":75.41,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.96,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.6,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.08,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"d5366350d0bb01b4","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_wee_patch16_reg1_gap_256.sbb_in1k","title":"ViT Wee/16 (256, reg1, GAP)","source":"timm/vit_wee_patch16_reg1_gap_256.sbb_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_wee_patch16_reg1_gap_256.sbb_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_wee_patch16_reg1_gap_256.sbb_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.456236,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.501323,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.457581,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.503296,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":2184,"unit":"qps","digits":1,"baseline":1989.22,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.51,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.54,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.67,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.15,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"d6178c15b42f5f84","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vit_xsmall_patch16_clip_224.tinyclip_yfcc15m","title":"ViT XSmall/16 (224, CLIP)","source":"timm/vit_xsmall_patch16_clip_224.tinyclip_yfcc15m","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvit_xsmall_patch16_clip_224.tinyclip_yfcc15m&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vit_xsmall_patch16_clip_224.tinyclip_yfcc15m","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.303348,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.304016,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":3280.7,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.6,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":76.8,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image"],"build":"75a6776ee9b01def","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"google/vit-base-patch16-224","title":"ViT-B/16","source":"google/vit-base-patch16-224","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=google%2Fvit-base-patch16-224&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"google/vit-base-patch16-224","hardware":"thor","precision":"FP8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.06763,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.40808,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Throughput","value":929.564,"unit":"qps","digits":1,"baseline":704.548,"baselineLabel":"Baseline TensorRT"}]},{"title":"Accuracy","subtitle":"ImageNette validation · 3,925 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":86.47133757961784,"unit":"%","digits":2,"baseline":85.98726114649682,"baselineLabel":"Baseline TensorRT","headline":true},{"description":"Top-5","value":98.06369426751593,"unit":"%","digits":2,"baseline":98.06369426751593,"baselineLabel":"Baseline TensorRT"}]},{"title":"Size","metrics":[{"description":"Engine size","value":85.9142,"unit":"MiB","higherIsBetter":false,"digits":1,"baseline":167.178,"baselineLabel":"Baseline TensorRT","improvementWord":"smaller","headline":true}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"torchvision/vit_b_16","title":"ViT-B/16 (torchvision)","source":"torchvision/vit_b_16","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fvit_b_16&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/vit_b_16","hardware":"orin-nano-super","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":6.32617,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":157.789,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNette validation · 3,925 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":91.36305732484077,"unit":"%","digits":2,"baseline":91.56687898089172,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":99.3375796178344,"unit":"%","digits":2,"baseline":99.43949044585987,"baselineLabel":"FP32 model"}]},{"title":"Size","metrics":[{"description":"Engine size","value":146,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/vit_b_16","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.4347441401332617,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":696.9883842195851,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":80.45400000000001,"unit":"%","digits":2,"baseline":81.072,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":95.024,"unit":"%","digits":2,"baseline":95.318,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/vit_b_16","hardware":"rtx-2080-ti","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.2675,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"p99 latency","value":1.278,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":788.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNette validation · 3,925 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":90.981,"unit":"%","digits":2,"headline":true},{"description":"Top-5","value":99.338,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Engine size","value":86.35,"unit":"MiB","higherIsBetter":false,"digits":2,"headline":true}]}],"labels":["Image","Classification"],"build":"202610132144614c","updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/vit_b_16","hardware":"rtx-2080-ti","variant":"torchvision_int8_embedl-deploy_tensorrt-rtx-2080-ti","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.2675,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"p99 latency","value":1.278,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":788.4,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNette validation · 3,925 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":90.981,"unit":"%","digits":2,"headline":true},{"description":"Top-5","value":99.338,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Engine size","value":86.35,"unit":"MiB","higherIsBetter":false,"digits":2,"headline":true}]}],"labels":["Image","Classification"],"build":"202610132144614c","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/vit_b_32","title":"ViT-B/32","source":"torchvision/vit_b_32","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fvit_b_32&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/vit_b_32","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.491723999788519,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":670.3652955518377,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":75.092,"unit":"%","digits":2,"baseline":75.912,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":92.034,"unit":"%","digits":2,"baseline":92.466,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/vit_l_16","title":"ViT-L/16","source":"torchvision","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fvit_l_16&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/vit_l_16","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":5.926,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":5.996,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":170.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":77.87,"unit":"%","headline":true},{"description":"Top-5","value":93.68,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"torchvision/vit_l_16","hardware":"l4","variant":"torchvision_int8_modelopt_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":5.926,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":5.996,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":170.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":77.87,"unit":"%","headline":true},{"description":"Top-5","value":93.68,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/vit_l_32","title":"ViT-L/32","source":"torchvision/vit_l_32","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fvit_l_32&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/vit_l_32","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":3.5573131404817104,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":281.1110409764449,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":76.18599999999999,"unit":"%","digits":2,"baseline":76.972,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":92.584,"unit":"%","digits":2,"baseline":93.07,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vitamin_large_224.datacomp1b_clip","title":"ViTamin Large (224, CLIP)","source":"timm/vitamin_large_224.datacomp1b_clip","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvitamin_large_224.datacomp1b_clip&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vitamin_large_224.datacomp1b_clip","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.00626,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":6.00419,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":5.15283,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":6.24445,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":199.7,"unit":"qps","digits":1,"baseline":166.508,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4.26,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.74,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"f6f6911c0ac258c9","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vitamin_large_256.datacomp1b_clip","title":"ViTamin Large (256, CLIP)","source":"timm/vitamin_large_256.datacomp1b_clip","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvitamin_large_256.datacomp1b_clip&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vitamin_large_256.datacomp1b_clip","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.78236,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":6.72796,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":5.92957,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":6.87201,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":172.9,"unit":"qps","digits":1,"baseline":148.599,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4.26,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.74,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"81926dd1201196a3","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vitamin_large_336.datacomp1b_clip","title":"ViTamin Large (336, CLIP)","source":"timm/vitamin_large_336.datacomp1b_clip","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvitamin_large_336.datacomp1b_clip&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vitamin_large_336.datacomp1b_clip","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.96329,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":10.3909,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":100.3,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4.27,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.75,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"83e3bc2e8745c36e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vitamin_large_384.datacomp1b_clip","title":"ViTamin Large (384, CLIP)","source":"timm/vitamin_large_384.datacomp1b_clip","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvitamin_large_384.datacomp1b_clip&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vitamin_large_384.datacomp1b_clip","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":20.3155,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":20.9668,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":49.2,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4.35,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.83,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"9f06ab68c972ab5e","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vitamin_large2_224.datacomp1b_clip","title":"ViTamin Large2 (224, CLIP)","source":"timm/vitamin_large2_224.datacomp1b_clip","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvitamin_large2_224.datacomp1b_clip&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vitamin_large2_224.datacomp1b_clip","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.07372,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":6.0224,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":5.20941,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":6.22223,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":197,"unit":"qps","digits":1,"baseline":166,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4.27,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.74,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"5c1a3ff9436a6073","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vitamin_large2_256.datacomp1b_clip","title":"ViTamin Large2 (256, CLIP)","source":"timm/vitamin_large2_256.datacomp1b_clip","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvitamin_large2_256.datacomp1b_clip&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vitamin_large2_256.datacomp1b_clip","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.94171,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":6.67509,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":6.16406,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":6.89148,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":168.3,"unit":"qps","digits":1,"baseline":149.776,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4.27,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.74,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"ba2e3fb72bc3113b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vitamin_large2_336.datacomp1b_clip","title":"ViTamin Large2 (336, CLIP)","source":"timm/vitamin_large2_336.datacomp1b_clip","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvitamin_large2_336.datacomp1b_clip&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vitamin_large2_336.datacomp1b_clip","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":9.88816,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":10.2209,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":101.1,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4.27,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.75,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"2cf69392c39357d0","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vitamin_large2_384.datacomp1b_clip","title":"ViTamin Large2 (384, CLIP)","source":"timm/vitamin_large2_384.datacomp1b_clip","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvitamin_large2_384.datacomp1b_clip&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vitamin_large2_384.datacomp1b_clip","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":20.412,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":21.3657,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":49,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":4.35,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":3.83,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"e0c423c527a8abae","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vitamin_small_224.datacomp1b_clip","title":"ViTamin Small (224, CLIP)","source":"timm/vitamin_small_224.datacomp1b_clip","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvitamin_small_224.datacomp1b_clip&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vitamin_small_224.datacomp1b_clip","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.08246,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.08582,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":922.6,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.77,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.25,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"8442561a03485a9b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vitamin_xlarge_256.datacomp1b_clip","title":"ViTamin XLarge (256, CLIP)","source":"timm/vitamin_xlarge_256.datacomp1b_clip","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvitamin_xlarge_256.datacomp1b_clip&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vitamin_xlarge_256.datacomp1b_clip","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.89052,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":9.15448,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":112.5,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":5.52,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":5,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"f503f676938aeb2b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/vitamin_xlarge_336.datacomp1b_clip","title":"ViTamin XLarge (336, CLIP)","source":"timm/vitamin_xlarge_336.datacomp1b_clip","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvitamin_xlarge_336.datacomp1b_clip&granularity=auto&v=54a3184f70d529074fbcbd6f&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vitamin_xlarge_336.datacomp1b_clip","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":19.8161,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":20.7724,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":50.5,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":5.52,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":5,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"8dd81be5a080b15c","updatedAt":"2026-11-06T15:36:58.000Z"}]},{"name":"timm/vitamin_xlarge_384.datacomp1b_clip","title":"ViTamin XLarge (384, CLIP)","source":"timm/vitamin_xlarge_384.datacomp1b_clip","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fvitamin_xlarge_384.datacomp1b_clip&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/vitamin_xlarge_384.datacomp1b_clip","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":24.7259,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":25.6306,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":40.4,"unit":"qps","digits":1}]},{"title":"Size","metrics":[{"description":"Total size, download","value":5.52,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":5,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image"],"build":"421f8dbcee67a118","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/wide_resnet101_2","title":"Wide ResNet-101-2","source":"torchvision/wide_resnet101_2","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fwide_resnet101_2&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/wide_resnet101_2","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.1971434578299522,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":835.3217765669356,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":80.862,"unit":"%","digits":2,"baseline":82.51,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":95.254,"unit":"%","digits":2,"baseline":96.02,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/wide_resnet101_2.tv2_in1k","title":"Wide ResNet-101-2 (timm)","source":"timm/wide_resnet101_2.tv2_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fwide_resnet101_2.tv2_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/wide_resnet101_2.tv2_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.921189,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.60808,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.924805,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.61322,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1083.8,"unit":"qps","digits":1,"baseline":621.308,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.66,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.96,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.93,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":1.41,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"fed10d316d9f7757","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"torchvision/wide_resnet50_2","title":"Wide ResNet-50-2","source":"torchvision/wide_resnet50_2","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=torchvision%2Fwide_resnet50_2&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"torchvision/wide_resnet50_2","hardware":"orin-nano-super","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":2.36511,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":422.369,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNette validation · 3,925 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":88.30573248407644,"unit":"%","digits":2,"baseline":88.73885350318471,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":99.28662420382166,"unit":"%","digits":2,"baseline":99.3375796178344,"baselineLabel":"FP32 model"}]},{"title":"Size","metrics":[{"description":"Engine size","value":66,"unit":"MiB","higherIsBetter":false,"digits":1,"headline":true}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"},{"model":"torchvision/wide_resnet50_2","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · GPU compute only · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":0.742299621924758,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"Throughput","value":1347.1649054690795,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet-1k validation · 50,000 images · FP32 model as baseline","metrics":[{"description":"Top-1","value":81.04400000000001,"unit":"%","digits":2,"baseline":81.602,"baselineLabel":"FP32 model","headline":true},{"description":"Top-5","value":95.48,"unit":"%","digits":2,"baseline":95.758,"baselineLabel":"FP32 model"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-10-29T16:07:23.000Z"}]},{"name":"timm/wide_resnet50_2.racm_in1k","title":"Wide ResNet-50-2 (timm)","source":"timm/wide_resnet50_2.racm_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fwide_resnet50_2.racm_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/wide_resnet50_2.racm_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.57572,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.00496,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.578064,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.00696,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1731.5,"unit":"qps","digits":1,"baseline":993.62,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.11,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.45,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.29,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.77,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"ef2f712f35a92364","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/legacy_xception.tf_in1k","title":"Xception (legacy)","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Flegacy_xception.tf_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/legacy_xception.tf_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":1.949,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"p99 latency","value":1.957,"unit":"ms","higherIsBetter":false,"digits":2},{"description":"Throughput","value":512.6,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":75.35,"unit":"%","headline":true},{"description":"Top-5","value":92.39,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"},{"model":"timm/legacy_xception.tf_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.89434,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.90594,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":527.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":75.37,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":92.4,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.81,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.29,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"f90395d2c89d47b2","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xception41p.ra3_in1k","title":"Xception-41-P","source":"timm/xception41p.ra3_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxception41p.ra3_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xception41p.ra3_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.649198,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.937319,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.650635,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":0.940125,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1536.5,"unit":"qps","digits":1,"baseline":1065.25,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.44,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.48,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.82,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.3,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"d69877b984459363","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xception65p.ra3_in1k","title":"Xception-65-P","source":"timm/xception65p.ra3_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxception65p.ra3_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xception65p.ra3_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.08086,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.38291,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":1.09436,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.38684,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":916.4,"unit":"qps","digits":1,"baseline":722.379,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.87,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.33,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.97,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.45,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"9b70550d1f2a9e96","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xcit_large_24_p16_224.fb_dist_in1k","title":"XCiT Large-24/16 (224)","source":"timm/xcit_large_24_p16_224.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxcit_large_24_p16_224.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xcit_large_24_p16_224.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.22476,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.80405,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":3.29211,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":3.88452,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":310,"unit":"qps","digits":1,"baseline":262.776,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.05,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.85,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.65,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.13,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"3709271339154fb9","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xcit_large_24_p16_384.fb_dist_in1k","title":"XCiT Large-24/16 (384)","source":"timm/xcit_large_24_p16_384.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxcit_large_24_p16_384.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xcit_large_24_p16_384.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":5.84494,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":6.71629,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":5.98663,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":6.87695,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":171,"unit":"qps","digits":1,"baseline":148.859,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":85.1,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.36,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.65,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.13,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"73b0c16c95e63a0a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xcit_large_24_p8_224.fb_dist_in1k","title":"XCiT Large-24/8 (224)","source":"timm/xcit_large_24_p8_224.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxcit_large_24_p8_224.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xcit_large_24_p8_224.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":7.87696,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":8.70977,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":8.09045,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":8.86157,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":126.9,"unit":"qps","digits":1,"baseline":114.794,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.6,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.26,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.65,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.12,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"7ad4712b3e797582","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xcit_large_24_p8_384.fb_dist_in1k","title":"XCiT Large-24/8 (384)","source":"timm/xcit_large_24_p8_384.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxcit_large_24_p8_384.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xcit_large_24_p8_384.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":22.0492,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":22.5348,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":45.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":85.4,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.58,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":2.65,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":2.12,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"43a8f14626ea9f17","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xcit_medium_24_p16_224.fb_dist_in1k","title":"XCiT Medium-24/16 (224)","source":"timm/xcit_medium_24_p16_224.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxcit_medium_24_p16_224.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xcit_medium_24_p16_224.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.16805,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.61742,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":2.2337,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":2.68365,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":460.9,"unit":"qps","digits":1,"baseline":381.856,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.88,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.7,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.47,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.95,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"38d5bc0608119af6","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xcit_medium_24_p16_384.fb_dist_in1k","title":"XCiT Medium-24/16 (384)","source":"timm/xcit_medium_24_p16_384.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxcit_medium_24_p16_384.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xcit_medium_24_p16_384.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.58987,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.16421,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":3.66003,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":4.26978,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":278.4,"unit":"qps","digits":1,"baseline":240.043,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.54,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.87,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.47,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.95,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"03a140cbdb0f9c64","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xcit_medium_24_p8_224.fb_dist_in1k","title":"XCiT Medium-24/8 (224)","source":"timm/xcit_medium_24_p8_224.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxcit_medium_24_p8_224.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xcit_medium_24_p8_224.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.45743,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.53796,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":224.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.73,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.47,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.95,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"92ff1dfd8e78cb48","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xcit_nano_12_p16_224.fb_dist_in1k","title":"XCiT Nano-12/16 (224)","source":"timm/xcit_nano_12_p16_224.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxcit_nano_12_p16_224.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xcit_nano_12_p16_224.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.600836,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.601379,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1660.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":68.74,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":88.54,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.56,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":38.1,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"a41d4fbaf131e14d","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xcit_nano_12_p8_224.fb_dist_in1k","title":"XCiT Nano-12/8 (224)","source":"timm","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxcit_nano_12_p8_224.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xcit_nano_12_p8_224.fb_dist_in1k","hardware":"l4","precision":"INT8","metrics":[{"title":"Latency","subtitle":"batch 1 · trtexec · Baseline TensorRT as baseline","metrics":[{"description":"Median latency","value":780.396,"unit":"µs","higherIsBetter":false,"digits":1,"headline":true},{"description":"p99 latency","value":781.677,"unit":"µs","higherIsBetter":false,"digits":1},{"description":"Throughput","value":1278.8,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation · 50,000 images · Baseline TensorRT as baseline","metrics":[{"description":"Top-1","value":59.97,"unit":"%","headline":true},{"description":"Top-5","value":79.12,"unit":"%"}]}],"labels":["Image","Classification"],"runnable":false,"updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xcit_small_12_p16_224.fb_dist_in1k","title":"XCiT Small-12/16 (224)","source":"timm/xcit_small_12_p16_224.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxcit_small_12_p16_224.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xcit_small_12_p16_224.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.878131,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.08182,"baselineLabel":"Baseline TensorRT FP16","headline":true},{"description":"GPU compute, p99","value":0.88031,"unit":"ms","higherIsBetter":false,"digits":3,"baseline":1.08557,"baselineLabel":"Baseline TensorRT FP16"},{"description":"Throughput","value":1136.8,"unit":"qps","digits":1,"baseline":923.048,"baselineLabel":"Baseline TensorRT FP16"}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":82.45,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.05,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.82,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.3,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"bec6add7ae84e0ee","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xcit_small_12_p16_384.fb_dist_in1k","title":"XCiT Small-12/16 (384)","source":"timm/xcit_small_12_p16_384.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxcit_small_12_p16_384.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xcit_small_12_p16_384.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.49596,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.49927,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":667.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.99,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.83,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.82,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.3,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"26ad523cc7e0578b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xcit_small_12_p8_224.fb_dist_in1k","title":"XCiT Small-12/8 (224)","source":"timm/xcit_small_12_p8_224.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxcit_small_12_p8_224.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xcit_small_12_p8_224.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.72228,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.75641,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":580.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.42,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.46,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.82,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.3,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"7f962def2984ae61","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xcit_small_12_p8_384.fb_dist_in1k","title":"XCiT Small-12/8 (384)","source":"timm/xcit_small_12_p8_384.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxcit_small_12_p8_384.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xcit_small_12_p8_384.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.60486,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.70844,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":217.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.36,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.92,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.82,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.3,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"70669e4ec483f29a","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xcit_small_24_p16_224.fb_dist_in1k","title":"XCiT Small-24/16 (224)","source":"timm/xcit_small_24_p16_224.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxcit_small_24_p16_224.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xcit_small_24_p16_224.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.95255,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.95514,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":511.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.07,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.39,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.06,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.54,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"1522d65e6b79d6ed","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xcit_small_24_p16_384.fb_dist_in1k","title":"XCiT Small-24/16 (384)","source":"timm/xcit_small_24_p16_384.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxcit_small_24_p16_384.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xcit_small_24_p16_384.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":2.76833,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.84418,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":361,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.37,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.98,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.06,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.54,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"aa1da3da28b59704","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xcit_small_24_p8_224.fb_dist_in1k","title":"XCiT Small-24/8 (224)","source":"timm/xcit_small_24_p8_224.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxcit_small_24_p8_224.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xcit_small_24_p8_224.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":3.36271,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":3.43402,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":297.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":84.24,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.99,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.06,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.54,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"7b6927b961791147","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xcit_small_24_p8_384.fb_dist_in1k","title":"XCiT Small-24/8 (384)","source":"timm/xcit_small_24_p8_384.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxcit_small_24_p8_384.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xcit_small_24_p8_384.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":8.62289,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":8.91528,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":115.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":85.14,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":97.37,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":1.06,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.54,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"7d711991ebfd4279","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xcit_tiny_12_p16_224.fb_dist_in1k","title":"XCiT Tiny-12/16 (224)","source":"timm/xcit_tiny_12_p16_224.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxcit_tiny_12_p16_224.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xcit_tiny_12_p16_224.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.693336,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.694244,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1438.7,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":74.36,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":91.66,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.6,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":80,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"e4d0ddbb23a737c0","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xcit_tiny_12_p16_384.fb_dist_in1k","title":"XCiT Tiny-12/16 (384)","source":"timm/xcit_tiny_12_p16_384.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxcit_tiny_12_p16_384.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xcit_tiny_12_p16_384.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.878556,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.88324,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1136.2,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":76.9,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":93.13,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.6,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":80.2,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"6895dcd50f5acbfd","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xcit_tiny_12_p8_224.fb_dist_in1k","title":"XCiT Tiny-12/8 (224)","source":"timm/xcit_tiny_12_p8_224.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxcit_tiny_12_p8_224.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xcit_tiny_12_p8_224.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":0.969203,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":0.971924,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":1030.1,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.99,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.34,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.6,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":80,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"7b30667c4ba1dfc7","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xcit_tiny_12_p8_384.fb_dist_in1k","title":"XCiT Tiny-12/8 (384)","source":"timm/xcit_tiny_12_p8_384.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxcit_tiny_12_p8_384.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xcit_tiny_12_p8_384.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.99643,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":2.02118,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":500.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.33,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.16,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.6,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":80.6,"unit":"MiB","higherIsBetter":false,"digits":1}]}],"labels":["Image","Classification"],"build":"aef6a159d047ab9b","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xcit_tiny_24_p16_224.fb_dist_in1k","title":"XCiT Tiny-24/16 (224)","source":"timm/xcit_tiny_24_p16_224.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxcit_tiny_24_p16_224.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xcit_tiny_24_p16_224.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.28704,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.28815,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":775.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":78.9,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":94.3,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.66,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.14,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"3c67c2c3e1c6b612","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xcit_tiny_24_p16_384.fb_dist_in1k","title":"XCiT Tiny-24/16 (384)","source":"timm/xcit_tiny_24_p16_384.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxcit_tiny_24_p16_384.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xcit_tiny_24_p16_384.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.62214,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.62842,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":615.9,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":80.92,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.44,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.66,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.14,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"3e111c748a201c49","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xcit_tiny_24_p8_224.fb_dist_in1k","title":"XCiT Tiny-24/8 (224)","source":"timm/xcit_tiny_24_p8_224.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxcit_tiny_24_p8_224.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xcit_tiny_24_p8_224.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":1.82197,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":1.8252,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":548.5,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":81.78,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":95.83,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.66,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.14,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"e3cc32b655c94024","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"timm/xcit_tiny_24_p8_384.fb_dist_in1k","title":"XCiT Tiny-24/8 (384)","source":"timm/xcit_tiny_24_p8_384.fb_dist_in1k","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=timm%2Fxcit_tiny_24_p8_384.fb_dist_in1k&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"timm/xcit_tiny_24_p8_384.fb_dist_in1k","hardware":"l4","variant":"int8_embedl_tensorrt","precision":"INT8","metrics":[{"title":"Latency","subtitle":"TensorRT 11.3.0.99, INT8, batch 1 (trtexec)","metrics":[{"description":"GPU compute, mean","value":4.2308,"unit":"ms","higherIsBetter":false,"digits":3,"headline":true},{"description":"GPU compute, p99","value":4.34387,"unit":"ms","higherIsBetter":false,"digits":3},{"description":"Throughput","value":236.3,"unit":"qps","digits":1}]},{"title":"Accuracy","subtitle":"ImageNet validation set, 50,000 images","metrics":[{"description":"Top-1, INT8 engine","value":83.05,"unit":"%","digits":2,"headline":true},{"description":"Top-5, INT8 engine","value":96.42,"unit":"%","digits":2}]},{"title":"Size","metrics":[{"description":"Total size, download","value":0.66,"unit":"GiB","higherIsBetter":false,"digits":2,"headline":true},{"description":"Container size, download","value":0.52,"unit":"GiB","higherIsBetter":false,"digits":2},{"description":"Model size, download","value":0.14,"unit":"GiB","higherIsBetter":false,"digits":2}]}],"labels":["Image","Classification"],"build":"18820ac11f8b2ac1","updatedAt":"2026-11-02T14:44:56.000Z"}]},{"name":"embedl/pi05-droid","upstream":"Physical-Intelligence/openpi","title":"π0.5 DROID","upstreamUrl":"https://github.com/Physical-Intelligence/openpi","source":"embedl/pi05-droid","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=lerobot%2Fpi05_droid&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"embedl/pi05-droid","hardware":"thor","precision":"FP8","metrics":[{"title":"Latency","subtitle":"One 15-action chunk · batch 1 · 10 denoising steps","metrics":[{"description":"Policy, end to end, median","value":82.3,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"Engine alone, median","value":55.06,"unit":"ms","higherIsBetter":false,"digits":2}]},{"title":"Accuracy","subtitle":"RoboLab-120 · 120 tasks × 10 episodes · published by RoboLab for pi05_droid_jointpos","metrics":[{"description":"Task success on RoboLab-120 (published)","value":28,"unit":"%","higherIsBetter":true,"digits":2,"headline":true},{"description":"Actions cosine vs OpenPI PyTorch, min","value":0.99912,"unit":"","higherIsBetter":true,"digits":4}]}],"labels":["Text","VLA","Robot Policy"],"build":"2026102018287bc7","updatedAt":"2026-10-29T16:07:23.000Z","simulatorReachable":false},{"model":"embedl/pi05-droid","hardware":"thor","variant":"fp8-nvfp4","precision":"FP8 + NVFP4","metrics":[{"title":"Latency","subtitle":"One 15-action chunk · batch 1 · 10 denoising steps","metrics":[{"description":"Policy, end to end, median","value":57.93,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"Engine alone, median","value":48.95,"unit":"ms","higherIsBetter":false,"digits":2}]},{"title":"Accuracy","subtitle":"RoboLab-120 · 120 tasks × 10 episodes · published by RoboLab for pi05_droid_jointpos","metrics":[{"description":"Task success on RoboLab-120 (published)","value":28,"unit":"%","higherIsBetter":true,"digits":2,"headline":true},{"description":"Actions cosine vs OpenPI PyTorch, min","value":0.998876,"unit":"","higherIsBetter":true,"digits":4}]}],"labels":["Text","VLA","Robot Policy"],"build":"202610172106ee6a","updatedAt":"2026-10-29T16:07:23.000Z","simulatorReachable":false},{"model":"embedl/pi05-droid","hardware":"h200","precision":"FP8","metrics":[{"title":"Latency","subtitle":"One 15-action chunk · batch 1 · 10 denoising steps","metrics":[{"description":"Policy, end to end, median","value":28.79,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true},{"description":"Engine alone, median","value":19.35,"unit":"ms","higherIsBetter":false,"digits":2}]},{"title":"Accuracy","subtitle":"RoboLab-120 · 120 tasks × 10 episodes · published by RoboLab for pi05_droid_jointpos","metrics":[{"description":"Task success on RoboLab-120 (published)","value":28,"unit":"%","higherIsBetter":true,"digits":2,"headline":true},{"description":"Actions cosine vs OpenPI PyTorch, min","value":0.99918,"unit":"","higherIsBetter":true,"digits":4}]}],"labels":["Text","VLA","Robot Policy"],"build":"202610172106e563","updatedAt":"2026-10-29T16:07:23.000Z","simulatorReachable":false}]},{"name":"embedl/pi05-libero","upstream":"Physical-Intelligence/openpi","title":"π0.5 LIBERO","upstreamUrl":"https://github.com/Physical-Intelligence/openpi","source":"embedl/pi05-libero","graphUrl":"https://hfviewer.com/api/graph-preview.svg?source=lerobot%2Fpi05_libero_finetuned_v044&granularity=auto&tracking_actor=internal&tracking_actor_label=embedl-models-graph","builds":[{"model":"embedl/pi05-libero","hardware":"thor","precision":"FP8 + FP16","metrics":[{"title":"Latency","subtitle":"One action chunk · batch 1 · 10 denoising steps · 30 runs","metrics":[{"description":"Latency per action chunk, median","value":66.85136950000015,"unit":"ms","higherIsBetter":false,"digits":2,"headline":true}]},{"title":"Accuracy","subtitle":"LIBERO · 65 paired episodes · OpenPI PyTorch BF16 as baseline","metrics":[{"description":"Task success","value":98.46153846153847,"unit":"%","higherIsBetter":true,"digits":2,"baseline":96.92307692307692,"baselineLabel":"OpenPI PyTorch BF16","headline":true},{"description":"Task success, held-out init states 2–4","value":97.43589743589743,"unit":"%","higherIsBetter":true,"digits":2,"baseline":97.43589743589743,"baselineLabel":"OpenPI PyTorch BF16"}]}],"labels":["Text","VLA","Robot Policy"],"build":"20261018164038ff","updatedAt":"2026-10-29T16:07:23.000Z"}]}],"available":true}