Models · Classification

FocalNet

FocalNet (Yang et al. 2022; tiny, small and base with small or large receptive field, ImageNet-1k): a classifier with focal modulation instead of attention. MIT.

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

  • Paper Focal Modulation Networks NeurIPS 2022 · first released 22 Mar 2022
  • Authors’ code microsoft/FocalNet (PyTorch), not the code forked here: timm carries the architecture and the converted authors' checkpoints in its own, simpler API.
Checkpoint

Paper 2022 Added in 0.2.0 Evaluated with 0.2.0

Portable, RTX 5060 Ti 16 GB 7.36 ms 224x224, fp16
vs the faster original 1.1× original fp32: 8.39 ms
Peak memory 95 MiB original: 157 MiB
Accuracy vs original fp32 +0.04 portable top-1 %, ImageNet-1k val (82.86 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("focalnet_tiny_srf.ms_in1k")  # kernels on a supported GPU, else portable PyTorch
result = model(image)
result.logits
result.backend  # what ran: mode, device, dtype, kernels
fastforks bench focalnet_tiny_srf.ms_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)

BackenddtypeTime per call (ms)vs original fp32Peak memory
allocated / reserved
Load (s)First call (s)
Original fp32fp32 8.39–157 MiB/178 MiB1.70.83
Original fp16torch.autocast(float16) around the callfp16 10.40.81×197 MiB/206 MiB0.40.45
Portable fp32fp32not measured ––––
Portablefp16 7.361.1×95 MiB/118 MiB1.50.49
Kernelsnot measured ––––
Portable + torch.compile 3.722.3×120 MiB/136 MiB13 (build)–
TensorRT fp16fp16 0.8649.7×72 MiBenginenot comparable52 (build)–

Measured on RTX 5060 Ti 16 GB, torch 2.14.1+cu130, 2026-11-11. 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, 2026-11-14Originalfp3218.7 755 MiB process–224x224
Portablefp1618.3 731 MiB processmatches224x224
Portablefp3218.0 720 MiB processmatches224x224
Kernelsfp32not implemented–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
MIT source
Training data

Our reading of the licences, not legal advice.

Weights

focalnet_tiny_srf.ms_in1k.safetensors 114 MB· hosted on Hugging Face

fastforks downloads and checks these files on first use (fastforks download focalnet_tiny_srf.ms_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 fp32fp3282.8696.28 reference 5000 2.19
Original fp16fp1682.8896.30 +0.02 5000 1.22
Portable fp32fp3282.8896.28 +0.02 5000 3.04
Portablefp1682.9096.28 +0.04 5000 1.03
Kernelsnot measured

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 6f0a70e1, 2026-11-17; portable: RTX 5060 Ti, torch 2.14.1+cu130, fastforks 7ba7307f, 2026-11-11.

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| <= 2.9e-06 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 0.0027 and p99 <= 0.0098 (absolute error of the output logits).

CheckToleranceMeasured
CPU fp32, max |diff|≤ 0.0000029 not recorded
GPU default dtypes, mean |diff|≤ 0.0027 0.00167
GPU default dtypes, p99 |diff|≤ 0.0098 0.00581

Recorded for focalnet_tiny_srf.ms_in1k on RTX 5060 Ti, torch 2.10.0+cu130, 2026-11-17, with these kernels: cudnn_haccfvm_convfvm_extrafvm_mbfvm_r3

Details

FocalNet (Yang et al. 2022): a hierarchical classifier that replaces self-attention by focal modulation (stacked depthwise convolutions with gated aggregation).

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 (95 MiB allocated)
Outputs
logits
Checkpoint
arch="focalnet_tiny_srf.ms_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

FocalNet (Yang et al. 2022; tiny, small and base with small or large receptive field, ImageNet-1k): a classifier with focal modulation instead of attention. MIT.

Commercial use: yes.

variantinputbackendtime (ms)peak memory (MiB, allocated / reserved)
focalnet_tiny_srf.ms_in1ksynthetic (224x224)original (fp32)
original
portable
kernels
portable + torch.compile
TensorRT fp16
8.39
10.4
7.36
–
3.72
0.864
157 / 178
197 / 206
95 / 118
–
120 / 136
159 / 188
focalnet_small_srf.ms_in1ksynthetic (224x224)original (fp32)
original
portable
kernels
portable + torch.compile
TensorRT fp16
15.9
20.1
14.0
–
6.90
1.49
239 / 252
320 / 336
136 / 152
–
161 / 190
234 / 256
focalnet_base_srf.ms_in1ksynthetic (224x224)original (fp32)
original
portable
kernels
portable + torch.compile
TensorRT fp16
17.1
20.0
14.2
–
6.94
1.74
390 / 414
542 / 558
211 / 234
–
234 / 272
401 / 430
focalnet_tiny_lrf.ms_in1ksynthetic (224x224)original (fp32)
original
portable
kernels
portable + torch.compile
TensorRT fp16
8.92
11.4
8.07
–
3.94
0.952
158 / 180
199 / 208
95 / 118
–
120 / 138
158 / 186
focalnet_small_lrf.ms_in1ksynthetic (224x224)original (fp32)
original
portable
kernels
portable + torch.compile
TensorRT fp16
16.8
22.0
15.8
–
7.28
1.70
241 / 254
323 / 338
136 / 152
–
162 / 192
265 / 286
focalnet_base_lrf.ms_in1ksynthetic (224x224)original (fp32)
original
portable
kernels
portable + torch.compile
TensorRT fp16
18.0
22.1
15.5
–
7.47
2.08
393 / 416
546 / 560
212 / 234
–
235 / 274
404 / 432
Details
  • Original code: https://github.com/huggingface/pytorch-image-models at 0df212b369a5
  • Code licence: Apache-2.0 (permissive); files:
  • Weights licence: MIT
  • Licence evidence: https://huggingface.co/timm/focalnet_tiny_srf.ms_in1k/blob/de75c3a73e9ed84fee554a7138d4bac119751645/README.md
  • Training data: ImageNet-1k (ILSVRC 2012 classification)
  • Kernel libraries: cudnn_hacc, fvm_conv, fvm_extra, fvm_mb, fvm_r3
  • Variants: focalnet_tiny_srf.ms_in1k (default of focalnet_tiny_srf), focalnet_small_srf.ms_in1k (default of focalnet_small_srf), focalnet_base_srf.ms_in1k (default of focalnet_base_srf), focalnet_tiny_lrf.ms_in1k (default of focalnet_tiny_lrf), focalnet_small_lrf.ms_in1k (default of focalnet_small_lrf), focalnet_base_lrf.ms_in1k (default of focalnet_base_lrf)
  • 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): focalnet_tiny_srf.ms_in1k 224x224 (0.1 GiB peak allocated, 0.1 reserved); focalnet_small_srf.ms_in1k 224x224 (0.1 GiB peak allocated, 0.1 reserved); focalnet_base_srf.ms_in1k 224x224 (0.2 GiB peak allocated, 0.2 reserved); focalnet_tiny_lrf.ms_in1k 224x224 (0.1 GiB peak allocated, 0.1 reserved); focalnet_small_lrf.ms_in1k 224x224 (0.1 GiB peak allocated, 0.1 reserved); focalnet_base_lrf.ms_in1k 224x224 (0.2 GiB peak allocated, 0.2 reserved)
  • focalnet_tiny_srf.ms_in1k 224x224, first call: portable + torch.compile 13 s (cold Inductor cache), TensorRT fp16 52 s (ONNX + engine).
  • focalnet_small_srf.ms_in1k 224x224, first call: portable + torch.compile 18 s (cold Inductor cache), TensorRT fp16 65 s (ONNX + engine).
  • focalnet_base_srf.ms_in1k 224x224, first call: portable + torch.compile 18 s (cold Inductor cache), TensorRT fp16 64 s (ONNX + engine).
  • focalnet_tiny_lrf.ms_in1k 224x224, first call: portable + torch.compile 13 s (cold Inductor cache), TensorRT fp16 50 s (ONNX + engine).
  • focalnet_small_lrf.ms_in1k 224x224, first call: portable + torch.compile 19 s (cold Inductor cache), TensorRT fp16 62 s (ONNX + engine).
  • focalnet_base_lrf.ms_in1k 224x224, first call: portable + torch.compile 20 s (cold Inductor cache), TensorRT fp16 64 s (ONNX + engine).
  • focalnet_tiny_srf.ms_in1k original 224x224: torch.autocast(float16) around the call
  • focalnet_small_srf.ms_in1k original 224x224: torch.autocast(float16) around the call
  • focalnet_base_srf.ms_in1k original 224x224: torch.autocast(float16) around the call
  • focalnet_tiny_lrf.ms_in1k original 224x224: torch.autocast(float16) around the call
  • focalnet_small_lrf.ms_in1k original 224x224: torch.autocast(float16) around the call
  • focalnet_base_lrf.ms_in1k original 224x224: torch.autocast(float16) around the call
  • focalnet_tiny_srf.ms_in1k load / first call: original 0.4 s / 0.445 s, portable 1.49 s / 0.495 s
  • focalnet_small_srf.ms_in1k load / first call: original 0.54 s / 0.023 s, portable 1.73 s / 0.49 s
  • focalnet_base_srf.ms_in1k load / first call: original 0.89 s / 0.052 s, portable 1.92 s / 0.487 s
  • focalnet_tiny_lrf.ms_in1k load / first call: original 0.36 s / 0.154 s, portable 2.56 s / 0.771 s
  • focalnet_small_lrf.ms_in1k load / first call: original 0.54 s / 0.025 s, portable 1.86 s / 0.687 s
  • focalnet_base_lrf.ms_in1k load / first call: original 0.91 s / 0.033 s, portable 1.89 s / 0.616 s
  • 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 fp32note
focalnet_tiny_srf.ms_in1kImageNet-1k val, 5000 random images, top-1 / top-5 %82.86 / 96.2882.88 / 96.3082.90 / 96.2882.88 / 96.28
focalnet_small_srf.ms_in1kImageNet-1k val, 5000 random images, top-1 / top-5 %84.34 / 96.8484.30 / 96.8484.44 / 96.7684.40 / 96.74
focalnet_base_srf.ms_in1kImageNet-1k val, 5000 random images, top-1 / top-5 %84.20 / 96.8684.18 / 96.8684.16 / 96.8484.16 / 96.84
focalnet_tiny_lrf.ms_in1kImageNet-1k val, 5000 random images, top-1 / top-5 %82.88 / 96.1282.86 / 96.1082.86 / 96.2082.82 / 96.20
focalnet_small_lrf.ms_in1kImageNet-1k val, 5000 random images, top-1 / top-5 %84.36 / 96.8084.34 / 96.8084.32 / 96.8084.38 / 96.80
focalnet_base_lrf.ms_in1kImageNet-1k val, 5000 random images, top-1 / top-5 %84.64 / 96.6484.64 / 96.6484.56 / 96.5684.56 / 96.58

Measured on portable: NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 7ba7307f, 2026-11-11; portable32: NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 6f0a70e1, 2026-11-17.

Parity

CPU, float32: max |diff| <= 2.9e-06 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 0.0027 and p99 <= 0.0098 (absolute error of the output logits).

variantkernels error vs original, mean / p99GPU, torchkernels used
focalnet_tiny_srf.ms_in1k0.00167 / 0.00581RTX 5060 Ti, torch 2.10.0+cu130cudnn_hacc@cudnn91501, fvm_conv@f1d064e539e1:sm_120, fvm_extra@3208ce525a91:sm_120, fvm_mb@720dd3f29bf7:sm_120, fvm_r3@f54eb890e23c:sm_120
focalnet_small_srf.ms_in1k0.00124 / 0.00466RTX 5060 Ti, torch 2.10.0+cu130cudnn_hacc@cudnn91501, fvm_conv@f1d064e539e1:sm_120, fvm_extra@3208ce525a91:sm_120, fvm_mb@720dd3f29bf7:sm_120, fvm_r3@f54eb890e23c:sm_120
focalnet_base_srf.ms_in1k0.00131 / 0.00626RTX 5060 Ti, torch 2.10.0+cu130cudnn_hacc@cudnn91501, fvm_conv@f1d064e539e1:sm_120, fvm_extra@3208ce525a91:sm_120, fvm_mb@720dd3f29bf7:sm_120, fvm_r3@f54eb890e23c:sm_120
focalnet_tiny_lrf.ms_in1k0.00128 / 0.00548RTX 5060 Ti, torch 2.10.0+cu130cudnn_hacc@cudnn91501, fvm_conv@f1d064e539e1:sm_120, fvm_extra@3208ce525a91:sm_120, fvm_mb@720dd3f29bf7:sm_120, fvm_r3@f54eb890e23c:sm_120
focalnet_small_lrf.ms_in1k0.00179 / 0.00799RTX 5060 Ti, torch 2.10.0+cu130cudnn_hacc@cudnn91501, fvm_conv@f1d064e539e1:sm_120, fvm_extra@3208ce525a91:sm_120, fvm_mb@720dd3f29bf7:sm_120, fvm_r3@f54eb890e23c:sm_120
focalnet_base_lrf.ms_in1k0.00106 / 0.0054RTX 5060 Ti, torch 2.10.0+cu130cudnn_hacc@cudnn91501, 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 (portable · float16 · NVIDIA GeForce RTX 5060 Ti).

Input

Top 5

  1. Egyptian cat 68%
  2. tiger cat 12%
  3. tabby 5.4%
  4. lynx 0.2%
  5. shower cap <0.1%
Chelsea the cat, input
Input · 100%

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