ConvNeXt
Meta's ConvNeXt (Liu et al. 2022; tiny, small, base, large; ImageNet-1k, and ImageNet-22k fine-tuned at 224 and 384): a convolutional classifier of 7x7 depthwise convolutions, LayerNorm and MLP blocks. Apache-2.0.
stable Fork of huggingface/pytorch-image-models@0df212b · backends: portable, kernels
- Paper A ConvNet for the 2020s CVPR 2022 · first released 10 Jan 2022
- Authors’ code facebookresearch/ConvNeXt (PyTorch), not the code forked here: timm ships a maintained re-implementation with the same Meta ImageNet-1k/22k weights and a uniform API.
Paper 2022 Added in 0.2.0 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("convnext_base.fb_in22k_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 convnext_base.fb_in22k_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 fp32 | Peak memory allocated / reserved | Load (s) | First call (s) |
|---|---|---|---|---|---|---|
| Original fp32timm 1.0.30 network, bicubic resize + centre crop + normalisation on the GPU inside the timed call | fp32 | 7.78 | – | 390 MiB/398 MiB | 1.0 | 0.01 |
| Original fp16torch.autocast(fp16) around the call; timm 1.0.30 network, bicubic resize + centre crop + normalisation on the GPU inside the timed call | fp16 | 11.3 | 0.69× | 541 MiB/566 MiB | 0.9 | 0.01 |
| Portable fp32 | fp32 | not measured | – | – | – | – |
| Portable | fp16 | 5.84 | 1.3× | 211 MiB/234 MiB | 2.1 | 0.27 |
| Kernels | fp16 | 2.01 | 3.9× | 208 MiB/236 MiB | 2.4 | 0.26 |
| Portable + torch.compile | 3.37 | 2.3× | 323 MiB/334 MiB | 11 (build) | – | |
| TensorRT fp16 | fp16 | 1.45 | 5.4× | 200 MiBenginenot comparable | 31 (build) | – |
Measured on RTX 5060 Ti 16 GB, torch 2.14.1+cu130, 2026-11-10 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
| 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 | 25.7 | 2,092 MiB process | – | 224x224 |
| Portable | fp16 | 24.9 | 2,308 MiB process | matches | 224x224 | |
| Portable | fp32 | 27.8 | 2,324 MiB process | matches | 224x224 | |
| Kernels | fp16 | not 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
- 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
convnext_base.fb_in22k_ft_in1k.safetensors 354 MB· hosted on Hugging Facefastforks downloads and checks these files on first use (fastforks download convnext_base.fb_in22k_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 | 86.42 | 97.96 | reference | 5000 | 3.87 |
| Original fp16 | fp16 | 86.48 | 97.96 | +0.06 | 5000 | 1.73 |
| Portable fp32 | fp32 | 86.34 | 97.94 | −0.08 | 5000 | 4.22 |
| Portable | fp16 | 86.26 | 97.94 | −0.16 | 5000 | 1.18 |
| Kernels | fp16 | 86.30 | 97.94 | −0.12 | 5000 | 0.943 |
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, portable, kernels: RTX 5060 Ti, torch 2.14.1+cu130, 2026-11-10; portable fp32: RTX 5060 Ti, torch 2.14.1+cu130, fastforks 6f0a70e1, 2026-11-17.
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| <= 1.7e-05 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 0.0033 and p99 <= 0.012 (absolute error of the output logits).
| Check | Tolerance | Measured |
|---|---|---|
| CPU fp32, max |diff| | ≤ 0.000017 | not recorded |
| GPU default dtypes, mean |diff| | ≤ 0.0033 | 0.00099 |
| GPU default dtypes, p99 |diff| | ≤ 0.012 | 0.00361 |
Recorded for convnext_base.fb_in22k_ft_in1k on RTX 5060 Ti, torch 2.14.1+cu130, 2026-11-10, with these kernels: fvm_convfvm_seg
Details
ConvNeXt (Liu et al. 2022): a pure convolutional network of 7x7 depthwise conv + LayerNorm + MLP blocks, Meta's ImageNet-1k and ImageNet-22k checkpoints.
- Input / output
- Input: one RGB image. Output: ImageNet-1k class logits.
- Input size
- fixed 224x224: the input is resized (short side 256) and centre-cropped to it; tested up to 224x224 (291 MiB allocated)
- Outputs
logits- Checkpoint
arch="convnext_base.fb_in22k_ft_in1k"- CUDA kernels
fvm_convfused convolutions for image classifiersfvm_segfused kernels for Mask2Former and OneFormer
- Apple (MPS)
- fp32, fp16
Full reference, generated from the repository
Meta's ConvNeXt (Liu et al. 2022; tiny, small, base, large; ImageNet-1k, and ImageNet-22k fine-tuned at 224 and 384): a convolutional classifier of 7x7 depthwise convolutions, LayerNorm and MLP blocks. Apache-2.0.
Commercial use: yes.
| variant | input | backend | time (ms) | peak memory (MiB, allocated / reserved) |
|---|---|---|---|---|
convnext_tiny.fb_in1k | synthetic (224x224) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 4.05 6.14 3.31 1.15 2.09 0.682 | 155 / 170 198 / 216 94 / 110 92 / 110 150 / 170 184 / 200 |
convnext_small.fb_in1k | synthetic (224x224) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 7.44 11.3 5.97 1.84 3.47 1.16 | 238 / 252 321 / 338 136 / 150 134 / 152 211 / 230 278 / 294 |
convnext_base.fb_in1k | synthetic (224x224) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 7.77 11.2 5.84 2.00 3.37 1.45 | 390 / 398 541 / 566 211 / 234 208 / 236 323 / 334 482 / 494 |
convnext_large.fb_in1k | synthetic (224x224) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 10.7 11.5 5.81 3.28 3.45 2.50 | 828 / 854 1166 / 1180 423 / 432 419 / 434 642 / 652 986 / 1006 |
convnext_tiny.fb_in22k_ft_in1k | synthetic (224x224) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 4.11 6.13 3.34 1.15 2.08 0.687 | 155 / 172 198 / 216 94 / 110 92 / 110 150 / 170 184 / 200 |
convnext_small.fb_in22k_ft_in1k | synthetic (224x224) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 7.51 11.3 5.84 1.84 3.50 1.17 | 238 / 252 321 / 338 136 / 150 134 / 152 211 / 230 278 / 294 |
convnext_base.fb_in22k_ft_in1k | synthetic (224x224) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 7.78 11.3 5.84 2.01 3.37 1.45 | 390 / 398 541 / 566 211 / 234 208 / 236 323 / 334 482 / 494 |
convnext_large.fb_in22k_ft_in1k | synthetic (224x224) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 10.7 11.5 5.88 3.28 3.44 2.50 | 828 / 854 1166 / 1180 423 / 432 419 / 434 642 / 652 986 / 1006 |
convnext_tiny.fb_in22k_ft_in1k_384 | synthetic (224x224) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 4.59 6.05 3.30 1.56 2.05 0.970 | 182 / 210 204 / 216 108 / 132 96 / 110 162 / 170 192 / 226 |
convnext_small.fb_in22k_ft_in1k_384 | synthetic (224x224) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 8.04 11.2 5.82 2.53 3.36 1.68 | 264 / 292 328 / 338 150 / 172 138 / 152 223 / 232 317 / 350 |
convnext_base.fb_in22k_ft_in1k_384 | synthetic (224x224) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 11.6 11.2 5.79 3.05 3.41 2.31 | 423 / 436 549 / 562 229 / 252 213 / 232 336 / 354 493 / 522 |
convnext_large.fb_in22k_ft_in1k_384 | synthetic (224x224) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 22.9 11.5 6.11 5.21 5.80 4.23 | 864 / 894 1177 / 1260 449 / 492 426 / 454 654 / 672 1001 / 1040 |
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/convnext_base.fb_in22k_ft_in1k/blob/67758fcbcf9f007ec7630bf8636b74a8f24e52a0/README.md
- Training data: ImageNet-21k / ImageNet-22k (full ImageNet, fall 2011 release), ImageNet-1k (ILSVRC 2012 classification)
convnext_tiny.fb_in1k: ImageNet-1k (ILSVRC 2012 classification)convnext_small.fb_in1k: ImageNet-1k (ILSVRC 2012 classification)convnext_base.fb_in1k: ImageNet-1k (ILSVRC 2012 classification)convnext_large.fb_in1k: ImageNet-1k (ILSVRC 2012 classification)
- Kernel libraries:
fvm_conv,fvm_seg - Variants:
convnext_tiny.fb_in1k,convnext_small.fb_in1k,convnext_base.fb_in1k,convnext_large.fb_in1k,convnext_tiny.fb_in22k_ft_in1k,convnext_small.fb_in22k_ft_in1k,convnext_base.fb_in22k_ft_in1k(default ofconvnext_base),convnext_large.fb_in22k_ft_in1k(default ofconvnext_large),convnext_tiny.fb_in22k_ft_in1k_384,convnext_small.fb_in22k_ft_in1k_384,convnext_base.fb_in22k_ft_in1k_384,convnext_large.fb_in22k_ft_in1k_384 - Outputs: logits; label_set:
imagenet-1k - Input size:
convnext_tiny.fb_in1k,convnext_small.fb_in1k,convnext_base.fb_in1k,convnext_large.fb_in1k,convnext_tiny.fb_in22k_ft_in1k,convnext_small.fb_in22k_ft_in1k,convnext_base.fb_in22k_ft_in1k,convnext_large.fb_in22k_ft_in1k: fixed 224x224: the input is resized (short side 256) and centre-cropped to itconvnext_tiny.fb_in22k_ft_in1k_384,convnext_small.fb_in22k_ft_in1k_384,convnext_base.fb_in22k_ft_in1k_384,convnext_large.fb_in22k_ft_in1k_384: fixed 384x384: the input is resized (short side 384) and centre-cropped to it
- Tested up to (portable, fp16, RTX 5060 Ti 16 GB):
convnext_tiny.fb_in1k224x224 (0.1 GiB peak allocated, 0.2 reserved);convnext_small.fb_in1k224x224 (0.2 GiB peak allocated, 0.2 reserved);convnext_base.fb_in1k224x224 (0.3 GiB peak allocated, 0.4 reserved);convnext_large.fb_in1k224x224 (0.6 GiB peak allocated, 0.7 reserved);convnext_tiny.fb_in22k_ft_in1k224x224 (0.1 GiB peak allocated, 0.2 reserved);convnext_small.fb_in22k_ft_in1k224x224 (0.2 GiB peak allocated, 0.2 reserved);convnext_base.fb_in22k_ft_in1k224x224 (0.3 GiB peak allocated, 0.4 reserved);convnext_large.fb_in22k_ft_in1k224x224 (0.6 GiB peak allocated, 0.7 reserved);convnext_tiny.fb_in22k_ft_in1k_384384x384 (0.1 GiB peak allocated, 0.2 reserved);convnext_small.fb_in22k_ft_in1k_384384x384 (0.2 GiB peak allocated, 0.2 reserved);convnext_base.fb_in22k_ft_in1k_384384x384 (0.3 GiB peak allocated, 0.4 reserved);convnext_large.fb_in22k_ft_in1k_384384x384 (0.6 GiB peak allocated, 0.7 reserved) convnext_tiny.fb_in1k224x224, first call: portable + torch.compile 8 s (cold Inductor cache), TensorRT fp16 27 s (ONNX + engine).convnext_small.fb_in1k224x224, first call: portable + torch.compile 11 s (cold Inductor cache), TensorRT fp16 30 s (ONNX + engine).convnext_base.fb_in1k224x224, first call: portable + torch.compile 11 s (cold Inductor cache), TensorRT fp16 32 s (ONNX + engine).convnext_large.fb_in1k224x224, first call: portable + torch.compile 11 s (cold Inductor cache), TensorRT fp16 34 s (ONNX + engine).convnext_tiny.fb_in22k_ft_in1k224x224, first call: portable + torch.compile 8 s (cold Inductor cache), TensorRT fp16 28 s (ONNX + engine).convnext_small.fb_in22k_ft_in1k224x224, first call: portable + torch.compile 10 s (cold Inductor cache), TensorRT fp16 31 s (ONNX + engine).convnext_base.fb_in22k_ft_in1k224x224, first call: portable + torch.compile 11 s (cold Inductor cache), TensorRT fp16 31 s (ONNX + engine).convnext_large.fb_in22k_ft_in1k224x224, first call: portable + torch.compile 11 s (cold Inductor cache), TensorRT fp16 35 s (ONNX + engine).convnext_tiny.fb_in22k_ft_in1k_384224x224, first call: portable + torch.compile 8 s (cold Inductor cache), TensorRT fp16 28 s (ONNX + engine).convnext_small.fb_in22k_ft_in1k_384224x224, first call: portable + torch.compile 11 s (cold Inductor cache), TensorRT fp16 32 s (ONNX + engine).convnext_base.fb_in22k_ft_in1k_384224x224, first call: portable + torch.compile 11 s (cold Inductor cache), TensorRT fp16 31 s (ONNX + engine).convnext_large.fb_in22k_ft_in1k_384224x224, first call: portable + torch.compile 11 s (cold Inductor cache), TensorRT fp16 33 s (ONNX + engine).convnext_tiny.fb_in1koriginal (fp32) 224x224: timm 1.0.30 network, bicubic resize + centre crop + normalisation on the GPU inside the timed callconvnext_tiny.fb_in1koriginal 224x224: torch.autocast(fp16) around the call; timm 1.0.30 network, bicubic resize + centre crop + normalisation on the GPU inside the timed callconvnext_small.fb_in1koriginal (fp32) 224x224: timm 1.0.30 network, bicubic resize + centre crop + normalisation on the GPU inside the timed callconvnext_small.fb_in1koriginal 224x224: torch.autocast(fp16) around the call; timm 1.0.30 network, bicubic resize + centre crop + normalisation on the GPU inside the timed callconvnext_base.fb_in1koriginal (fp32) 224x224: timm 1.0.30 network, bicubic resize + centre crop + normalisation on the GPU inside the timed callconvnext_base.fb_in1koriginal 224x224: torch.autocast(fp16) around the call; timm 1.0.30 network, bicubic resize + centre crop + normalisation on the GPU inside the timed callconvnext_large.fb_in1koriginal (fp32) 224x224: timm 1.0.30 network, bicubic resize + centre crop + normalisation on the GPU inside the timed callconvnext_large.fb_in1koriginal 224x224: torch.autocast(fp16) around the call; timm 1.0.30 network, bicubic resize + centre crop + normalisation on the GPU inside the timed callconvnext_tiny.fb_in22k_ft_in1koriginal (fp32) 224x224: timm 1.0.30 network, bicubic resize + centre crop + normalisation on the GPU inside the timed callconvnext_tiny.fb_in22k_ft_in1koriginal 224x224: torch.autocast(fp16) around the call; timm 1.0.30 network, bicubic resize + centre crop + normalisation on the GPU inside the timed callconvnext_small.fb_in22k_ft_in1koriginal (fp32) 224x224: timm 1.0.30 network, bicubic resize + centre crop + normalisation on the GPU inside the timed callconvnext_small.fb_in22k_ft_in1koriginal 224x224: torch.autocast(fp16) around the call; timm 1.0.30 network, bicubic resize + centre crop + normalisation on the GPU inside the timed callconvnext_base.fb_in22k_ft_in1koriginal (fp32) 224x224: timm 1.0.30 network, bicubic resize + centre crop + normalisation on the GPU inside the timed callconvnext_base.fb_in22k_ft_in1koriginal 224x224: torch.autocast(fp16) around the call; timm 1.0.30 network, bicubic resize + centre crop + normalisation on the GPU inside the timed callconvnext_large.fb_in22k_ft_in1koriginal (fp32) 224x224: timm 1.0.30 network, bicubic resize + centre crop + normalisation on the GPU inside the timed callconvnext_large.fb_in22k_ft_in1koriginal 224x224: torch.autocast(fp16) around the call; timm 1.0.30 network, bicubic resize + centre crop + normalisation on the GPU inside the timed callconvnext_tiny.fb_in22k_ft_in1k_384original (fp32) 224x224: timm 1.0.30 network, bicubic resize + centre crop + normalisation on the GPU inside the timed callconvnext_tiny.fb_in22k_ft_in1k_384original 224x224: torch.autocast(fp16) around the call; timm 1.0.30 network, bicubic resize + centre crop + normalisation on the GPU inside the timed callconvnext_small.fb_in22k_ft_in1k_384original (fp32) 224x224: timm 1.0.30 network, bicubic resize + centre crop + normalisation on the GPU inside the timed callconvnext_small.fb_in22k_ft_in1k_384original 224x224: torch.autocast(fp16) around the call; timm 1.0.30 network, bicubic resize + centre crop + normalisation on the GPU inside the timed callconvnext_base.fb_in22k_ft_in1k_384original (fp32) 224x224: timm 1.0.30 network, bicubic resize + centre crop + normalisation on the GPU inside the timed callconvnext_base.fb_in22k_ft_in1k_384original 224x224: torch.autocast(fp16) around the call; timm 1.0.30 network, bicubic resize + centre crop + normalisation on the GPU inside the timed callconvnext_large.fb_in22k_ft_in1k_384original (fp32) 224x224: timm 1.0.30 network, bicubic resize + centre crop + normalisation on the GPU inside the timed callconvnext_large.fb_in22k_ft_in1k_384original 224x224: torch.autocast(fp16) around the call; timm 1.0.30 network, bicubic resize + centre crop + normalisation on the GPU inside the timed callconvnext_tiny.fb_in1kload / first call: original 0.41 s / 0.491 s, portable 1.48 s / 0.242 s, kernels 1.5 s / 0.245 sconvnext_small.fb_in1kload / first call: original 0.65 s / 0.014 s, portable 1.81 s / 0.219 s, kernels 1.59 s / 0.186 sconvnext_base.fb_in1kload / first call: original 0.95 s / 0.015 s, portable 1.91 s / 0.25 s, kernels 1.99 s / 0.27 sconvnext_large.fb_in1kload / first call: original 1.89 s / 0.016 s, portable 2.79 s / 0.251 s, kernels 2.58 s / 0.244 sconvnext_tiny.fb_in22k_ft_in1kload / first call: original 0.35 s / 0.008 s, portable 1.49 s / 0.241 s, kernels 1.45 s / 0.237 sconvnext_small.fb_in22k_ft_in1kload / first call: original 0.52 s / 0.013 s, portable 1.67 s / 0.222 s, kernels 1.93 s / 0.206 sconvnext_base.fb_in22k_ft_in1kload / first call: original 0.89 s / 0.014 s, portable 2.07 s / 0.251 s, kernels 1.97 s / 0.261 sconvnext_large.fb_in22k_ft_in1kload / first call: original 1.91 s / 0.015 s, portable 2.55 s / 0.25 s, kernels 2.62 s / 0.278 sconvnext_tiny.fb_in22k_ft_in1k_384load / first call: original 0.35 s / 0.01 s, portable 1.57 s / 0.243 s, kernels 1.48 s / 0.242 sconvnext_small.fb_in22k_ft_in1k_384load / first call: original 0.52 s / 0.013 s, portable 1.7 s / 0.217 s, kernels 1.81 s / 0.364 sconvnext_base.fb_in22k_ft_in1k_384load / first call: original 0.87 s / 0.014 s, portable 2.54 s / 0.248 s, kernels 1.91 s / 0.257 sconvnext_large.fb_in22k_ft_in1k_384load / first call: original 1.75 s / 0.016 s, portable 2.58 s / 0.253 s, kernels 2.51 s / 0.251 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.
| variant | dataset | original fp32 | original fp16 | portable | portable fp32 | kernels | note |
|---|---|---|---|---|---|---|---|
convnext_tiny.fb_in1k | ImageNet-1k val, 5000 random images, top-1 / top-5 % | 82.32 / 96.40 | 82.28 / 96.40 | 82.34 / 96.28 | 82.30 / 96.28 | 82.28 / 96.32 | |
convnext_small.fb_in1k | ImageNet-1k val, 5000 random images, top-1 / top-5 % | 83.66 / 96.64 | 83.68 / 96.64 | 83.64 / 96.56 | 83.62 / 96.56 | 83.66 / 96.56 | |
convnext_base.fb_in1k | ImageNet-1k val, 5000 random images, top-1 / top-5 % | 84.66 / 96.88 | 84.64 / 96.86 | 84.58 / 96.82 | 84.56 / 96.80 | 84.54 / 96.82 | |
convnext_large.fb_in1k | ImageNet-1k val, 5000 random images, top-1 / top-5 % | 85.44 / 96.78 | 85.40 / 96.78 | 85.54 / 96.78 | 85.56 / 96.80 | 85.56 / 96.84 | |
convnext_tiny.fb_in22k_ft_in1k | ImageNet-1k val, 5000 random images, top-1 / top-5 % | 83.64 / 96.82 | 83.66 / 96.82 | 83.46 / 96.78 | 83.46 / 96.80 | 83.44 / 96.80 | |
convnext_small.fb_in22k_ft_in1k | ImageNet-1k val, 5000 random images, top-1 / top-5 % | 84.78 / 97.66 | 84.76 / 97.66 | 84.96 / 97.66 | 84.98 / 97.66 | 84.96 / 97.66 | |
convnext_base.fb_in22k_ft_in1k | ImageNet-1k val, 5000 random images, top-1 / top-5 % | 86.42 / 97.96 | 86.48 / 97.96 | 86.26 / 97.94 | 86.34 / 97.94 | 86.30 / 97.94 | |
convnext_large.fb_in22k_ft_in1k | ImageNet-1k val, 5000 random images, top-1 / top-5 % | 87.54 / 98.26 | 87.54 / 98.26 | 87.38 / 98.26 | 87.36 / 98.26 | 87.38 / 98.26 | |
convnext_tiny.fb_in22k_ft_in1k_384 | ImageNet-1k val, 5000 random images, top-1 / top-5 % | 84.98 / 97.40 | 84.96 / 97.40 | 85.00 / 97.22 | 84.94 / 97.22 | 84.94 / 97.22 | |
convnext_small.fb_in22k_ft_in1k_384 | ImageNet-1k val, 5000 random images, top-1 / top-5 % | 86.36 / 98.04 | 86.36 / 98.04 | 86.16 / 98.18 | 86.14 / 98.18 | 86.16 / 98.18 | |
convnext_base.fb_in22k_ft_in1k_384 | ImageNet-1k val, 5000 random images, top-1 / top-5 % | 87.48 / 98.50 | 87.52 / 98.50 | 87.34 / 98.40 | 87.32 / 98.40 | 87.32 / 98.40 | |
convnext_large.fb_in22k_ft_in1k_384 | ImageNet-1k val, 5000 random images, top-1 / top-5 % | 88.36 / 98.50 | 88.34 / 98.50 | 88.08 / 98.50 | 88.06 / 98.50 | 88.08 / 98.50 |
Measured on portable: NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, 2026-11-10; portable32: NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 6f0a70e1, 2026-11-17; kernels: NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, 2026-11-10.
Parity
CPU, float32: max |diff| <= 1.7e-05 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 0.0033 and p99 <= 0.012 (absolute error of the output logits).
| variant | kernels error vs original, mean / p99 | GPU, torch | kernels used |
|---|---|---|---|
convnext_tiny.fb_in1k | 0.00089 / 0.00297 | RTX 5060 Ti, torch 2.14.1+cu130 | fvm_conv@f1d064e539e1:sm_120, fvm_seg@739693dff3b4:sm_120 |
convnext_small.fb_in1k | 0.00062 / 0.00245 | RTX 5060 Ti, torch 2.14.1+cu130 | fvm_conv@f1d064e539e1:sm_120, fvm_seg@739693dff3b4:sm_120 |
convnext_base.fb_in1k | 0.00079 / 0.00318 | RTX 5060 Ti, torch 2.14.1+cu130 | fvm_conv@f1d064e539e1:sm_120, fvm_seg@739693dff3b4:sm_120 |
convnext_large.fb_in1k | 0.00071 / 0.00254 | RTX 5060 Ti, torch 2.14.1+cu130 | fvm_conv@f1d064e539e1:sm_120, fvm_seg@739693dff3b4:sm_120 |
convnext_tiny.fb_in22k_ft_in1k | 0.00215 / 0.00696 | RTX 5060 Ti, torch 2.14.1+cu130 | fvm_conv@f1d064e539e1:sm_120, fvm_seg@739693dff3b4:sm_120 |
convnext_small.fb_in22k_ft_in1k | 0.00086 / 0.00284 | RTX 5060 Ti, torch 2.14.1+cu130 | fvm_conv@f1d064e539e1:sm_120, fvm_seg@739693dff3b4:sm_120 |
convnext_base.fb_in22k_ft_in1k | 0.00099 / 0.00361 | RTX 5060 Ti, torch 2.14.1+cu130 | fvm_conv@f1d064e539e1:sm_120, fvm_seg@739693dff3b4:sm_120 |
convnext_large.fb_in22k_ft_in1k | 0.00111 / 0.00377 | RTX 5060 Ti, torch 2.14.1+cu130 | fvm_conv@f1d064e539e1:sm_120, fvm_seg@739693dff3b4:sm_120 |
convnext_tiny.fb_in22k_ft_in1k_384 | 0.00216 / 0.00769 | RTX 5060 Ti, torch 2.14.1+cu130 | fvm_conv@f1d064e539e1:sm_120, fvm_seg@739693dff3b4:sm_120 |
convnext_small.fb_in22k_ft_in1k_384 | 0.00141 / 0.00525 | RTX 5060 Ti, torch 2.14.1+cu130 | fvm_conv@f1d064e539e1:sm_120, fvm_seg@739693dff3b4:sm_120 |
convnext_base.fb_in22k_ft_in1k_384 | 0.00102 / 0.00417 | RTX 5060 Ti, torch 2.14.1+cu130 | fvm_conv@f1d064e539e1:sm_120, fvm_seg@739693dff3b4:sm_120 |
convnext_large.fb_in22k_ft_in1k_384 | 0.00107 / 0.00383 | RTX 5060 Ti, torch 2.14.1+cu130 | fvm_conv@f1d064e539e1:sm_120, fvm_seg@739693dff3b4: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).