Models · Classification

FastViT

FastViT (Vasu et al. 2023): a hybrid classifier with reparameterisable token mixing.

stable Fork of huggingface/pytorch-image-models@0df212b · backends: portable, kernels

Checkpoint

Paper 2023 Added in 0.1.0 Evaluated with 0.2.0

Kernels, RTX 5060 Ti 16 GB 1.41 ms 256x256, fp16
vs the faster original 5.0× original fp32: 7.10 ms
Peak memory 80 MiB original: 129 MiB
Accuracy vs original fp32 +0.02 kernels top-1 %, ImageNet-1k val (83.78 original)

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("fastvit_sa24.apple_dist_in1k")  # kernels on a supported GPU, else portable PyTorch
result = model(image)
result.logits
result.backend  # what ran: mode, device, dtype, kernels
fastforks bench fastvit_sa24.apple_dist_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, 256x256

BackenddtypeTime per call (ms)vs original fp32Peak memory
allocated / reserved
Load (s)First call (s)
Original fp32timm FastViT with authors' reparameterized inference recipe; fusion is outside the timed callfp32 7.10–129 MiB/142 MiB0.30.02
Original fp16torch.autocast(float16) around the call; timm FastViT with authors' reparameterized inference recipe; fusion is outside the timed callfp16 10.90.65×157 MiB/176 MiB0.30.03
Portable fp32fp32not measured ––––
Portablefp16 6.031.2×81 MiB/110 MiB2.50.82
Kernelsfp16 1.415.0×80 MiB/132 MiB2.80.05
Portable + torch.compile 3.971.8×113 MiB/148 MiB8 (build)–
TensorRT fp16fp16 0.8188.7×72 MiBenginenot comparable55 (build)–

Measured on RTX 5060 Ti 16 GB, torch 2.14.1+cu130, 2026-11-11 to 2026-11-16. 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

DeviceBackenddtypems / callPeak memoryOutput vs referenceInput
Apple M1 Pro 16 GB MPS, torch 2.14.1, 2026-11-14Originalfp3224.8 566 MiB processmatches256x256
Portablefp1616.9 722 MiB processmatches256x256
Portablefp3215.8 665 MiB processmatches256x256
Kernelsfp16not implemented–256x256
NVIDIA L4 CUDA, torch 2.14.1+cu130, 2026-11-07Portablefp1612.6 57 MiB/88 MiB–224x224
Kernelsfp161.76 63 MiB/128 MiB–224x224
NVIDIA L40S CUDA, torch 2.14.1+cu130, 2026-11-07Portablefp169.55 57 MiB/88 MiB–224x224
Kernelsfp161.27 63 MiB/128 MiB–224x224
NVIDIA A10 CUDA, torch 2.14.1+cu130, 2026-11-07Portablefp166.57 57 MiB/88 MiB–224x224
Kernelsfp161.73 63 MiB/128 MiB–224x224
NVIDIA A100-SXM4-40GB CUDA, torch 2.14.1+cu130, 2026-11-07Portablefp168.28 57 MiB/88 MiB–224x224
Kernelsfp162.49 63 MiB/128 MiB–224x224
NVIDIA RTX PRO 6000 CUDA, torch 2.14.1+cu130, 2026-11-08Portablefp162.88 81 MiB/128 MiB–224x224
Kernelsfp161.31 111 MiB/188 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
LicenseRef-Apple-Sample-Code source
Training data
  • The weights are under Apple's sample-code licence (the timm Hugging Face card says only 'license: other'); that licence grants no patent rights.

Our reading of the licences, not legal advice.

Weights

fastvit_sa24.apple_dist_in1k.safetensors 86.6 MB· hosted on Hugging Face

fastforks downloads and checks these files on first use (fastforks download fastvit_sa24.apple_dist_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 %

Rundtypetop-1 % ↑top-5 % ↑Δ top-1 % vs fp32 originalimagesms / image
Original fp32fp3283.7896.48 reference 5000 2.05
Original fp16fp1683.8096.46 +0.02 5000 1.52
Portable fp32fp3283.7896.44 ±0.00 5000 2.37
Portablefp1683.7896.44 ±0.00 5000 1.08
Kernelsfp1683.8096.42 +0.02 5000 0.597

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, 2026-11-11; portable fp32: RTX 5060 Ti, torch 2.14.1+cu130, fastforks 7c91993, 2026-10-31; portable: RTX 5060 Ti, torch 2.14.1+cu130, fastforks 95e72cc0, 2026-11-11; 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).

CheckToleranceMeasured
CPU fp32, max |diff|≤ 0.002 not recorded
GPU default dtypes, mean |diff|≤ 0.02 0.00727
GPU default dtypes, p99 |diff|≤ 0.08 0.01865

Recorded for fastvit_sa24.apple_dist_in1k on RTX 5060 Ti, torch 2.14.1+cu130, 2026-11-04, with these kernels: cudnn_haccfvm_convfvm_extrafvm_mbfvm_r3

Details

FastViT (Vasu et al. 2023): reparameterisable RepMixer token mixing, attention in the last stage.

Input / output
Input: one RGB image. Output: ImageNet-1k class logits.
Input size
fixed 256x256: the input is resized (short side 284) and centre-cropped to it; tested up to 256x256 (81 MiB allocated)
Outputs
logits
Checkpoint
arch="fastvit_sa24.apple_dist_in1k"
CUDA kernels
  • cudnn_haccconvolutions with fp16 accumulation (cuDNN)
  • fvm_convfused convolutions for image classifiers
  • fvm_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

FastViT (Vasu et al. 2023): a hybrid classifier with reparameterisable token mixing.

Commercial use: yes.

variantinputbackendtime (ms)peak memory (MiB, allocated / reserved)
fastvit_sa24.apple_dist_in1ksynthetic (256x256)original (fp32)
original
portable
kernels
portable + torch.compile
TensorRT fp16
7.10
10.9
6.03
1.41
3.97
0.818
129 / 142
157 / 176
81 / 110
80 / 132
113 / 148
148 / 200
fastvit_t8.apple_dist_in1ksynthetic (256x256)original (fp32)
original
portable
kernels
portable + torch.compile
TensorRT fp16
3.07
4.91
3.25
0.709
2.55
0.368
56 / 72
57 / 76
45 / 72
45 / 76
77 / 92
78 / 108
fastvit_t12.apple_dist_in1ksynthetic (256x256)original (fp32)
original
portable
kernels
portable + torch.compile
TensorRT fp16
3.61
5.66
3.27
0.781
2.32
0.431
74 / 80
77 / 84
54 / 76
52 / 76
86 / 94
89 / 114
Details
  • Original code: https://github.com/huggingface/pytorch-image-models at 0df212b369a5
  • Code licence: Apache-2.0 (permissive); files:
  • Weights licence: LicenseRef-Apple-Sample-Code
  • Licence evidence: https://huggingface.co/timm/fastvit_sa24.apple_dist_in1k/blob/0b86f2b98f83274e9f03b6ddbcb3942b84efb839/README.md
  • Training data: ImageNet-1k (ILSVRC 2012 classification)
  • Licence note: The weights are under Apple's sample-code licence (the timm Hugging Face card says only 'license: other'); that licence grants no patent rights.
  • Kernel libraries: cudnn_hacc, fvm_conv, fvm_extra, fvm_mb, fvm_r3
  • Variants: fastvit_sa24.apple_dist_in1k (default of fastvit_sa24), fastvit_t8.apple_dist_in1k, fastvit_t12.apple_dist_in1k
  • Outputs: logits; label_set: imagenet-1k
  • Input size: fixed 256x256: the input is resized (short side 284) and centre-cropped to it
  • Tested up to (portable, fp16, RTX 5060 Ti 16 GB): fastvit_sa24.apple_dist_in1k 256x256 (0.1 GiB peak allocated, 0.1 reserved); fastvit_t8.apple_dist_in1k 256x256 (0.0 GiB peak allocated, 0.1 reserved); fastvit_t12.apple_dist_in1k 256x256 (0.1 GiB peak allocated, 0.1 reserved)
  • fastvit_sa24.apple_dist_in1k 256x256, first call: portable + torch.compile 8 s (cold Inductor cache), TensorRT fp16 54 s (ONNX + engine).
  • fastvit_t8.apple_dist_in1k 256x256, first call: portable + torch.compile 7 s (cold Inductor cache), TensorRT fp16 31 s (ONNX + engine).
  • fastvit_t12.apple_dist_in1k 256x256, first call: portable + torch.compile 7 s (cold Inductor cache), TensorRT fp16 40 s (ONNX + engine).
  • fastvit_sa24.apple_dist_in1k original (fp32) 256x256: timm FastViT with authors' reparameterized inference recipe; fusion is outside the timed call
  • fastvit_sa24.apple_dist_in1k original 256x256: torch.autocast(float16) around the call; timm FastViT with authors' reparameterized inference recipe; fusion is outside the timed call
  • fastvit_t8.apple_dist_in1k original (fp32) 256x256: timm FastViT with authors' reparameterized inference recipe; fusion is outside the timed call
  • fastvit_t8.apple_dist_in1k original 256x256: torch.autocast(float16) around the call; timm FastViT with authors' reparameterized inference recipe; fusion is outside the timed call
  • fastvit_t12.apple_dist_in1k original (fp32) 256x256: timm FastViT with authors' reparameterized inference recipe; fusion is outside the timed call
  • fastvit_t12.apple_dist_in1k original 256x256: torch.autocast(float16) around the call; timm FastViT with authors' reparameterized inference recipe; fusion is outside the timed call
  • fastvit_sa24.apple_dist_in1k load / first call: original 0.31 s / 0.028 s (256x256), portable 2.45 s / 0.818 s (256x256), kernels 2.77 s / 0.046 s (256x256)
  • fastvit_t8.apple_dist_in1k load / first call: original 0.21 s / 0.976 s (256x256), portable 2.1 s / 0.851 s (256x256), kernels 1.83 s / 0.047 s (256x256)
  • fastvit_t12.apple_dist_in1k load / first call: original 0.14 s / 0.708 s (256x256), portable 1.81 s / 0.406 s (256x256), kernels 2.01 s / 0.044 s (256x256)
  • Protocol, first call, 3 warm-up, peak reset, median of 3 rounds x median of 5

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.

variantdatasetoriginal fp32original fp16portableportable fp32kernelsnote
fastvit_sa24.apple_dist_in1kImageNet-1k val, 5000 random images, top-1 / top-5 %83.78 / 96.4883.80 / 96.4683.78 / 96.4483.78 / 96.4483.80 / 96.42
fastvit_t8.apple_dist_in1kImageNet-1k val, 5000 random images, top-1 / top-5 %78.76 / 93.3878.68 / 93.3278.66 / 93.3478.76 / 93.3678.66 / 93.38
fastvit_t12.apple_dist_in1kImageNet-1k val, 5000 random images, top-1 / top-5 %81.12 / 94.9881.08 / 94.9281.02 / 94.9681.16 / 94.9881.08 / 94.96

Measured on portable: NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 95e72cc0, 2026-11-11; portable32: NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 6f0a70e1 / 7c91993, 2026-10-31 to 2026-11-17; kernels: NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 939adfe8 / d4cb0d31, 2026-11-06 to 2026-11-11.

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).

variantkernels error vs original, mean / p99GPU, torchkernels used
fastvit_sa24.apple_dist_in1k0.00727 / 0.01865RTX 5060 Ti, torch 2.14.1+cu130cudnn_hacc@cudnn92400, fvm_conv@f1d064e539e1:sm_120, fvm_extra@3208ce525a91:sm_120, fvm_mb@720dd3f29bf7:sm_120, fvm_r3@f54eb890e23c:sm_120
fastvit_t8.apple_dist_in1k0.01525 / 0.04597RTX 5060 Ti, torch 2.14.1+cu130cudnn_hacc@cudnn92400, fvm_conv@f1d064e539e1:sm_120, fvm_extra@3208ce525a91:sm_120, fvm_mb@720dd3f29bf7:sm_120, fvm_r3@f54eb890e23c:sm_120
fastvit_t12.apple_dist_in1k0.02097 / 0.03606RTX 5060 Ti, torch 2.14.1+cu130cudnn_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).

Input

Top 5

  1. Egyptian cat 62%
  2. tiger cat 16%
  3. tabby 12%
  4. lynx 0.2%
  5. bow tie <0.1%
Chelsea the cat, input
Input · 100%

Input: Chelsea the cat, Stéfan van der Walt (CC0).