SHViT (fastforks)

SHViT (Yun and Ro 2024): a small classifier with a 16x16 convolutional stem and single-head attention on a part of the channels.

This repo is the model card only: it holds no weights. fastforks downloads the authors' weights from the links below and checks them against the SHA-256 in the model card. Code: https://github.com/kleinicke/fastforks. Kernels for the kernels backend: https://huggingface.co/fastforks/kernels. Original code: https://github.com/huggingface/pytorch-image-models at 0df212b369a5.

Author weight repos: https://huggingface.co/timm/shvit_s1.in1k, https://huggingface.co/timm/shvit_s2.in1k, https://huggingface.co/timm/shvit_s3.in1k, https://huggingface.co/timm/shvit_s4.in1k.

Task: classification. Tier: experimental. Outputs: logits. Backends: portable.

Input: one RGB image. Output: ImageNet-1k class logits.

Variants

Weights are the authors' files (not hosted here). SHA-256 shortened to 12 hex digits; the full hashes are in card.toml.

variant weights (author link) SHA-256 weights licence commercial use
shvit_s1.in1k@1 shvit_s1.in1k.safetensors ba604812c795 MIT yes
shvit_s2.in1k@1 shvit_s2.in1k.safetensors 7a8861fe040a MIT yes
shvit_s3.in1k@1 shvit_s3.in1k.safetensors 1d34510f1925 MIT yes
shvit_s4.in1k@1 shvit_s4.in1k.safetensors 5be4002f0bb6 MIT yes

Licence

Original-code strings, recorded as they are (research-use package; check them before any other use). Code: Apache-2.0.

Training data: ImageNet-1k (ILSVRC 2012 classification).

Licence evidence: https://huggingface.co/timm/shvit_s1.in1k/blob/d5c0c27b0d05c630284a3946882590e4054f1108/README.md

Input size

  • shvit_s1.in1k, shvit_s2.in1k, shvit_s3.in1k: fixed 224x224: the input is resized (short side 256) and centre-cropped to it
  • shvit_s4.in1k: fixed 256x256: the input is resized (short side 292) and centre-cropped to it Tested up to (portable, fp16, RTX 5060 Ti 16 GB): shvit_s1.in1k 224x224 (0.0 GiB peak allocated, 0.1 reserved); shvit_s2.in1k 224x224 (0.1 GiB peak allocated, 0.1 reserved); shvit_s3.in1k 224x224 (0.1 GiB peak allocated, 0.1 reserved); shvit_s4.in1k 256x256 (0.1 GiB peak allocated, 0.1 reserved)

Speed and memory

Classification, synthetic

model input time (ms): original fp32 original portable kernels kernels speed-up: vs original vs portable peak memory (MiB, allocated / reserved): original portable kernels
shvit_s1.in1k 224x224 8.41 11.1 6.75 – – – 70 / 76 47 / 58 –
shvit_s2.in1k 224x224 8.37 11.1 6.76 – – – 99 / 120 56 / 88 –
shvit_s3.in1k 224x224 9.40 12.5 8.11 – – – 115 / 134 62 / 112 –
shvit_s4.in1k 256x256 11.5 14.6 9.31 – – – 129 / 146 67 / 118 –

Startup and first call

model load (s): original portable kernels first call (s): original portable kernels
shvit_s1.in1k 0.31 2.16 – 0.92 0.41 –
shvit_s2.in1k 0.20 2.32 – 0.30 0.48 –
shvit_s3.in1k 0.21 2.80 – 0.57 0.54 –
shvit_s4.in1k 0.23 (256x256) 2.49 (256x256) – 0.31 (256x256) 0.44 (256x256) –

torch.compile and TensorRT

model input time: portable compile TensorRT fp16 kernels preparation: compile (cold first call) ONNX + engine build peak MiB (alloc / reserved): portable compile kernels TensorRT (engine + torch)
shvit_s1.in1k 224x224 6.75 ms 3.11 ms 0.510 ms – 10.2 s 23.3 s 47 / 58 78 / 98 – 77 / 90
shvit_s2.in1k 224x224 6.76 ms 3.90 ms 0.533 ms – 10.1 s 24.0 s 56 / 88 88 / 128 – 87 / 120
shvit_s3.in1k 224x224 8.11 ms 3.64 ms 0.575 ms – 10.9 s 24.6 s 62 / 112 94 / 132 – 92 / 144
shvit_s4.in1k 256x256 9.31 ms 4.11 ms 0.658 ms – 11.3 s 26.3 s 67 / 118 99 / 138 – 101 / 154

Fixed-size cells: time (ms) / peak allocated / peak reserved memory (MiB).

Hardware, protocol and caveats: benchmarks (RTX 5060 Ti 16 GB, sm_120). Original = the original code named above at the precision of portable (fp16, or (fp32) where fp16 is unsupported or not finite); original fp32 = its default fp32 run. Peak memory is the torch allocator's.

Accuracy

On public datasets

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 note
shvit_s1.in1k ImageNet-1k val, 5000 random images, top-1 / top-5 % 74.26 / 91.10 74.22 / 91.16 74.26 / 91.00
shvit_s2.in1k ImageNet-1k val, 5000 random images, top-1 / top-5 % 76.74 / 92.40 76.82 / 92.40 76.82 / 92.44
shvit_s3.in1k ImageNet-1k val, 5000 random images, top-1 / top-5 % 78.06 / 93.48 78.04 / 93.44 77.96 / 93.50
shvit_s4.in1k ImageNet-1k val, 5000 random images, top-1 / top-5 % 80.38 / 94.50 80.38 / 94.52 80.40 / 94.52

Measured on NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 7ba7307f, 2026-11-11.

Parity with the original code

Every variant is checked against reference outputs of the original code at its pinned commit. CPU, float32: max |diff| <= 1.8e-05 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 0.006 and p99 <= 0.023 (absolute error of the output logits).

Usage

from fastforks import classification

model = classification.load("shvit_s1.in1k")  # mode="auto": the preferred backend from recorded measurements
result = model("photo.jpg")
result.top(5)  # top-5 classes
result.backend  # what ran, e.g. Backend(mode='portable', device='cuda:0', dtype='float16')

Install with pip install fastforks (see https://github.com/kleinicke/fastforks). Other variants: shvit_s1.in1k, shvit_s2.in1k, shvit_s3.in1k, shvit_s4.in1k.

Kernels backend

No kernels backend: not built yet: convolutions with BatchNorm plus single-head attention on feature maps need a dedicated kernels runner.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Collection including fastforks/shvit