Skip to content
Log in

torchvision/swin_t

Embedl-optimized
Swin-T architecture graph
Swin-T architecture graph
  • torchvision/swin_t
  • INT8
  • L4
Labels
  • Image
  • Classification
Build
Embedl-optimized
Precision
INT8
Runtime
TensorRT 10.8.0.43 with Embedl optimizations
Input size
224×224
Run Swin-T on L4

Swin-T is an image classifier from torchvision. It labels an image with one of the 1,000 ImageNet classes.

Methodology

  • Latency: trtexec on the NVIDIA L4 with TensorRT 10.8.0.43: batch 1, CUDA graphs, median GPU compute time (no copies to and from the GPU).
  • Accuracy: the ImageNette validation split, 3,925 images, on both engines.
  • Memory: peak device memory while the engine runs.
  • Size: the serialized TensorRT engine.
  • Baseline TensorRT: the original model in FP16 on upstream TensorRT 10.8.0.43, same settings, same device.
  • Build: Embedl Deploy fusion and quantization, calibrated on 128 ImageNette training images, exported to ONNX and compiled with trtexec.