Skip to content

Instantly share code, notes, and snippets.

@rwightman
Created July 13, 2022 05:22
Show Gist options
  • Save rwightman/6c37e408192f2a94d29023099445e13f to your computer and use it in GitHub Desktop.
Save rwightman/6c37e408192f2a94d29023099445e13f to your computer and use it in GitHub Desktop.
timm vit models, eager vs aot vs torchscript, AMP, PyTorch 1.12
model infer_samples_per_sec infer_step_time infer_batch_size infer_img_size train_samples_per_sec train_step_time train_batch_size train_img_size param_count
vit_small_patch16_224 2444.7 104.691 256 224 955.88 267.078 256 224 22.05
vit_relpos_medium_patch16_224 1107.38 231.158 256 224 502.75 253.69 128 224 38.75
vit_base_patch16_224 1013.88 252.477 256 224 358.36 356.433 128 224 86.57
vit_base_patch16_384 288.27 888.045 256 384 102.82 300.795 31 384 86.86
model infer_samples_per_sec infer_step_time infer_batch_size infer_img_size train_samples_per_sec train_step_time train_batch_size train_img_size param_count
vit_small_patch16_224 2740.74 93.382 256 224 1004.66 253.975 256 224 22.05
vit_relpos_medium_patch16_224 1654.63 154.695 256 224 610.1 418.453 256 224 38.75
vit_base_patch16_224 1081.47 236.697 256 224 370.73 344.446 128 224 86.57
vit_large_patch16_224 350.32 730.746 256 224 112.18 568.934 64 224 304.33
vit_base_patch16_384 277.05 923.987 256 384 93.2 342.512 32 384 86.86
model infer_samples_per_sec infer_step_time infer_batch_size infer_img_size train_samples_per_sec train_step_time train_batch_size train_img_size param_count
vit_small_patch16_224 2964.42 86.341 256 224 1035.12 246.505 256 224 22.05
vit_relpos_medium_patch16_224 1891.39 135.333 256 224 608.34 209.313 128 224 38.75
vit_base_patch16_224 1131.94 226.139 256 224 378.65 337.283 128 224 86.57
vit_large_patch16_224 366.05 699.342 256 224 104.09 305.97 32 224 304.33
vit_base_patch16_384 330.34 774.932 256 384 104.41 219.514 23 384 86.86
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment