Vision Transformer
Vision Transformer (Dosovitskiy et al. 2020): a transformer classifier on 16x16 image patches.
stable Fork of huggingface/pytorch-image-models@0df212b · backends: portable, kernels
- Paper An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale ICLR 2021 · first released 22 Oct 2020
- Paper How to train your ViT? Data, Augmentation, and Regularization in Vision Transformers TMLR 2022 · first released 18 Jun 2021
- Authors’ code google-research/vision_transformer (JAX), not the code forked here: The authors' code is JAX; timm's PyTorch port carries the converted AugReg weights (further fine-tuned by timm).
Checkpoint vit_base_patch16_224.augreg2_in21k_ft_in1k
Paper 2020 Added in 0.1.0 Evaluated with 0.2.0
Use it
pip install fastforks Lean install with only what this model needs: pip install "fastforks-core[timm]"
from fastforks import classification
model = classification.load("vit_base_patch16_224.augreg2_in21k_ft_in1k") # kernels on a supported GPU, else portable PyTorch
result = model(image)
result.logits
result.backend # what ran: mode, device, dtype, kernels fastforks bench vit_base_patch16_224.augreg2_in21k_ft_in1k Times the model on this machine. load(…, mode="portable") or mode="kernels" picks a backend; usage guide.
Speed and memory
Every backend on one GPU (RTX 5060 Ti 16 GB) and the same input, the authors’ code included. Kernels runs fp16. Portable runs fp16 on CUDA.
Input: synthetic, 224x224 (the task’s standard size)
| Backend | dtype | Time per call (ms) | vs original fp16 | Peak memory allocated / reserved | Load (s) | First call (s) |
|---|---|---|---|---|---|---|
| Original fp32 | fp32 | 6.33 | 0.87× | 371 MiB/386 MiB | 0.8 | 0.15 |
| Original fp16torch.autocast(fp16) around the forward | fp16 | 5.48 | – | 531 MiB/548 MiB | 2.5 | 0.28 |
| Portable fp32 | fp32 | not measured | – | – | – | – |
| Portable | fp16 | 3.02 | 1.8× | 205 MiB/390 MiB | 0.7 | 0.00 |
| Kernels | fp16 | 1.66 | 3.3× | 200 MiB/248 MiB | 2.2 | 0.07 |
| Portable + torch.compile | 2.10 | 2.6× | 238 MiB/266 MiB | 7 (build) | – | |
| TensorRT fp16 | fp16 | 1.14 | 4.8× | 196 MiBenginenot comparable | 29 (build) | – |
Measured on RTX 5060 Ti 16 GB, torch 2.14.1+cu130, 2026-11-03 to 2026-11-15. First call, 3 warm-up, peak reset, median of 3 rounds x median of 5. Ratios divide two
times of this table (same GPU, same input). Original fp16 is the authors’ own precision option, else torch.autocast. torch.compile and TensorRT: build time instead of load time. TensorRT memory (engine, not comparable) is the GPU memory the engine took, read from the driver because TensorRT allocates outside PyTorch, so it can't be compared with the PyTorch peaks of the other rows.
Other devices
| Device | Backend | dtype | ms / call | Peak memory | Output vs reference | Input |
|---|---|---|---|---|---|---|
| Apple M1 Pro 16 GB MPS, torch 2.14.1, 2026-11-14 | Original | fp32 | 17.4 | 1,258 MiB process | matches | 224x224 |
| Portable | fp16 | 18.7 | 1,497 MiB process | matches | 224x224 | |
| Portable | fp32 | 16.9 | 1,314 MiB process | matches | 224x224 | |
| Kernels | fp16 | not implemented | – | 224x224 | ||
| NVIDIA L4 CUDA, torch 2.14.1+cu130, 2026-11-07 | Portable | fp16 | 4.92 | 179 MiB/206 MiB | – | 224x224 |
| Kernels | fp16 | 1.92 | 252 MiB/308 MiB | – | 224x224 | |
| NVIDIA L40S CUDA, torch 2.14.1+cu130, 2026-11-07 | Portable | fp16 | 3.51 | 179 MiB/206 MiB | – | 224x224 |
| Kernels | fp16 | 1.22 | 252 MiB/308 MiB | – | 224x224 | |
| NVIDIA A10 CUDA, torch 2.14.1+cu130, 2026-11-07 | Portable | fp16 | 2.85 | 179 MiB/206 MiB | – | 224x224 |
| Kernels | fp16 | 1.77 | 252 MiB/308 MiB | – | 224x224 | |
| NVIDIA A100-SXM4-40GB CUDA, torch 2.14.1+cu130, 2026-11-08 | Portable | fp16 | 3.52 | 179 MiB/206 MiB | – | 224x224 |
| Kernels | fp16 | 2.21 | 252 MiB/308 MiB | – | 224x224 | |
| NVIDIA RTX PRO 6000 CUDA, torch 2.14.1+cu130, 2026-11-08 | Portable | fp16 | 1.16 | 203 MiB/226 MiB | – | 224x224 |
| Kernels | fp16 | 1.02 | 300 MiB/368 MiB | – | 224x224 | |
CUDA memory is torch’s peak allocated / reserved; Apple has no such counter, so the Mac rows show the process footprint (weights, Python and torch included). “Output vs reference” compares the device’s output with the stored golden output under the card’s tolerance; – = not checked on that device. Kernels “not implemented” = no kernels for this model on that backend yet: it runs portable there.
Licence and weights
Commercial use allowed
- Code
- Apache-2.0
- Weights
- Apache-2.0 source
- Training data
-
- ImageNet-21k / ImageNet-22k (full ImageNet, fall 2011 release) noncommercial, non-commercial
- ImageNet-1k (ILSVRC 2012 classification) noncommercial, non-commercial
Our reading of the licences, not legal advice.
Weights
vit_base_patch16_224.augreg2_in21k_ft_in1k.safetensors 346 MB· hosted on Hugging Facefastforks downloads and checks these files on first use (fastforks download vit_base_patch16_224.augreg2_in21k_ft_in1k).
Accuracy
The test set through the authors’ code and through ours: same checkpoint, same protocol. The distance between the original’s fp32 and fp16 runs is what half precision alone costs; our runs should stay within it.
ImageNet-1k val
5000 random images, top-1 / top-5 %
| Run | dtype | top-1 % ↑ | top-5 % ↑ | Δ top-1 % vs fp32 original | images | ms / image |
|---|---|---|---|---|---|---|
| Original fp32 | fp32 | 85.88 | 97.70 | reference | 5000 | 3.72 |
| Original fp16 | fp16 | 85.90 | 97.70 | +0.02 | 5000 | 1.25 |
| Portable fp32 | fp32 | 85.86 | 97.68 | −0.02 | 5000 | 3.89 |
| Portable | fp16 | 85.84 | 97.70 | −0.04 | 5000 | 0.933 |
| Kernels | fp16 | 85.88 | 97.68 | ±0.00 | 5000 | 0.809 |
fastforks: the model's own resize / crop on the GPU; original: the library's pretrained loader and its PIL eval transform. Original = the authors’ code, same checkpoint and protocol. Measured on original fp32, original fp16: RTX 5060 Ti, torch 2.14.1+cu130, 2026-11-06; portable fp32: RTX 5060 Ti, torch 2.14.1+cu130, fastforks 7c91993, 2026-10-31; portable, kernels: RTX 5060 Ti, torch 2.14.1+cu130, fastforks 939adfe8, 2026-11-06.
Closeness to the original
Beyond the test-set scores: the raw output of ours against the original’s fp32 output on the same input, pixel by pixel. CPU, float32: max |diff| <= 0.002 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 0.02 and p99 <= 0.08 (absolute error of the output logits).
| Check | Tolerance | Measured |
|---|---|---|
| CPU fp32, max |diff| | ≤ 0.002 | not recorded |
| GPU default dtypes, mean |diff| | ≤ 0.02 | 0.00273 |
| GPU default dtypes, p99 |diff| | ≤ 0.08 | 0.01018 |
Recorded for vit_base_patch16_224.augreg2_in21k_ft_in1k on RTX 5060 Ti, torch 2.14.1+cu130, 2026-11-04, with these kernels: cudnn_haccfvm_convfvm_extrafvm_mbfvm_r3
Details
Vision Transformer (Dosovitskiy et al. 2020): 16x16 patches, pre-norm transformer encoder, class token.
- Input / output
- Input: one RGB image. Output: ImageNet-1k class logits.
- Input size
- fixed 224x224: the input is resized (short side 248) and centre-cropped to it; tested up to 224x224 (214 MiB allocated)
- Outputs
logits- Checkpoint
arch="vit_base_patch16_224.augreg2_in21k_ft_in1k"- CUDA kernels
cudnn_haccconvolutions with fp16 accumulation (cuDNN)fvm_convfused convolutions for image classifiersfvm_extrafused transformer row operations (rotary split, SwiGLU with LayerNorm)fvm_mbfused MBConv blocks (MobileNet and EfficientNet families)fvm_r3further fused classifier blocks
- Apple (MPS)
- fp32, fp16
Full reference, generated from the repository
Vision Transformer (Dosovitskiy et al. 2020): a transformer classifier on 16x16 image patches.
Commercial use: yes.
| variant | input | backend | time (ms) | peak memory (MiB, allocated / reserved) |
|---|---|---|---|---|
vit_base_patch16_224.augreg2_in21k_ft_in1k | synthetic (224x224) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 6.33 5.48 3.02 1.66 2.10 1.14 | 371 / 386 531 / 548 205 / 390 200 / 248 238 / 266 394 / 420 |
Details
- Original code: https://github.com/huggingface/pytorch-image-models at
0df212b369a5 - Code licence: Apache-2.0 (permissive); files:
- Weights licence: Apache-2.0
- Licence evidence: https://huggingface.co/timm/vit_base_patch16_224.augreg2_in21k_ft_in1k/blob/063c6c38a5d8510b2e57df480445e94b231dad2c/README.md
- Training data: ImageNet-21k / ImageNet-22k (full ImageNet, fall 2011 release), ImageNet-1k (ILSVRC 2012 classification)
- Kernel libraries:
cudnn_hacc,fvm_conv,fvm_extra,fvm_mb,fvm_r3 - Variants:
vit_base_patch16_224.augreg2_in21k_ft_in1k(default ofvit_base_patch16_224) - Outputs: logits; label_set:
imagenet-1k - Input size: fixed 224x224: the input is resized (short side 248) and centre-cropped to it
- Tested up to (portable, fp16, RTX 5060 Ti 16 GB):
vit_base_patch16_224.augreg2_in21k_ft_in1k224x224 (0.2 GiB peak allocated, 0.3 reserved) vit_base_patch16_224.augreg2_in21k_ft_in1k224x224, first call: portable + torch.compile 7 s (cold Inductor cache), TensorRT fp16 29 s (ONNX + engine).vit_base_patch16_224.augreg2_in21k_ft_in1koriginal 224x224: torch.autocast(fp16) around the forwardvit_base_patch16_224.augreg2_in21k_ft_in1kload / first call: original 2.49 s / 0.283 s, portable 0.73 s / 0.004 s, kernels 2.05 s / 0.069 s- Protocol, median of 5 after 3 warm-up, uint8 CUDA input (synthetic for bench_inputs sizes), card defaults; gc before each load; peak = max over the first call and the timed calls
Measured
fastforks: the model's own resize / crop on the GPU; original: the library's pretrained loader and its PIL eval transform. Original = the original code named under Details, same checkpoint and protocol.
| variant | dataset | original fp32 | original fp16 | portable | portable fp32 | kernels | note |
|---|---|---|---|---|---|---|---|
vit_base_patch16_224.augreg2_in21k_ft_in1k | ImageNet-1k val, 5000 random images, top-1 / top-5 % | 85.88 / 97.70 | 85.90 / 97.70 | 85.84 / 97.70 | 85.86 / 97.68 | 85.88 / 97.68 |
Measured on portable: NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 939adfe8, 2026-11-06; portable32: NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 7c91993, 2026-10-31; kernels: NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 939adfe8, 2026-11-06.
Parity
CPU, float32: max |diff| <= 0.002 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 0.02 and p99 <= 0.08 (absolute error of the output logits).
| variant | kernels error vs original, mean / p99 | GPU, torch | kernels used |
|---|---|---|---|
vit_base_patch16_224.augreg2_in21k_ft_in1k | 0.00273 / 0.01018 | RTX 5060 Ti, torch 2.14.1+cu130 | cudnn_hacc@cudnn92400, fvm_conv@f1d064e539e1:sm_120, fvm_extra@3208ce525a91:sm_120, fvm_mb@720dd3f29bf7:sm_120, fvm_r3@f54eb890e23c:sm_120 |
Example output
One fixed input through this checkpoint, the same for every model of the task (kernels · float16 · NVIDIA GeForce RTX 5060 Ti).
Top 5
Input: Chelsea the cat, Stéfan van der Walt (CC0).