Skip to content
Log in

torchvision/vit_b_16

Embedl Deploy · TensorRT
vit_b_16 architecture graph
vit_b_16 architecture graph
  • torchvision/vit_b_16
  • INT8
  • RTX 2080 Ti
Labels
  • Image
  • Classification
Build
Embedl-optimized
Precision
INT8
Runtime
TensorRT 11.3.0.99 with Embedl optimizations
Input size
224×224
Run ViT-B/16 on RTX 2080 Ti

ViT-B/16 is an image classifier from torchvision. It labels an image with one of the 1,000 ImageNet classes.

Methodology

  • Latency: on the NVIDIA GeForce RTX 2080 Ti: gPU time of one engine execution, batch 1 (copies excluded), 23572 runs.
  • Accuracy: ImageNette validation (ImageNet labels), 3925 images, on the NVIDIA GeForce RTX 2080 Ti.
  • Size: the serialized TensorRT engine.
  • Build: Embedl Deploy INT8 (int8 symmetric activations; int8 symmetric per-channel weights), calibrated on 32 ImageNette train images, from torchvision ViT_B_16_Weights.IMAGENET1K_V1.

Files

Image Size

Model

docker.io/embedl/models:torchvision.vit_b_16.int8.rtx-2080-ti.202610132144614c

 —

Runtime

docker.io/embedl/models:runtimes.tensorrt.amd64.2026101815014ee0

 —