Morph MORPH
SIGN IN SIGN UP
huggingface / pytorch-image-models UNCLAIMED

The largest collection of PyTorch image encoders / backbones. Including train, eval, inference, export scripts, and pretrained weights -- ResNet, ResNeXT, EfficientNet, NFNet, Vision Transformer (ViT), MobileNetV4, MobileNet-V3 & V2, RegNet, DPN, CSPNet, Swin Transformer, MaxViT, CoAtNet, ConvNeXt, and more

0 0 31 Python

COMMITS

main
May 27, 2026
R
Remove some changes to existing behaviour and small correctness issues with optimizer. tensor step/LR support.
Ross Wightman committed
fbe27d6
R
Update README.md
Ross Wightman committed
0a0b4f4
R
Use timm/ hub locations
Ross Wightman committed
50d0fc2
R
Initial EUPE model defs, slight tweaks for weight conversion. Fix #2690
Ross Wightman committed
7377f08
R
Add tipsv2 vision encoders
Ross Wightman committed
08fa5cd
May 26, 2026
R
Support better torch.compile use by using torch.tensor for step count in more timm optimizers, enhancing some shared helpers, tweaking some capturable use cases for optimizers that support it.
Ross Wightman committed
29993ff
R
Update version to 1.0.28.dev0
Ross Wightman committed
9c0aa42
May 20, 2026
J
validate.py: add --seed for reproducible validation
Jan Bernloehr committed
d45cb11
May 8, 2026
R
Update README.md for release
Ross Wightman committed
0e968e1
R
Version 1.0.27
Ross Wightman committed
6a49560
April 29, 2026
R
Fix empty command line kwarg passing
Ross Wightman committed
6a928a1
April 24, 2026
R
Avoid test triggering weights_only error for old PyTorch
Ross Wightman committed
a60ce8b
R
A few more tweaks to task setup for module access. Add back legacy_train.py as alternative.
Ross Wightman committed
7dff294
R
Missed _helpers for task based resume/checkpoint
Ross Wightman committed
1953da3
R
More task fixups, made 'trainable_module' consistent, base optimization off that, move ema into task.
Ross Wightman committed
8bdb7ae
April 23, 2026
R
Fix some performance regressions with torch.compile + Tasks. Fix #2693
Ross Wightman committed
56e3981
R
Add encoder_pool option to gemma4 classification model to toggle soft tokens from encoder vs full sequence
Ross Wightman committed
6ce166b
R
Update README.md
Ross Wightman committed
eb11943
R
Final model rename to better mach vision arch vs parent VLM arch, push to hub
Ross Wightman committed
3d8763c
April 22, 2026
R
Switch to timm RmsNorm to avoid gap w/ old PyTorch versions that timm still supports
Ross Wightman committed
b293674
R
Fix unit tests for gemma4 vit
Ross Wightman committed
f7419a3
R
Add patchify channels_last flag, though only ended up using it for batch. Gemma4 ViT interface cleanup, split into encoder / classifier models as optimal pooling / norm differs. Test fine-tune.
Ross Wightman committed
7766c22
April 21, 2026
R
More formatting tweaks, ran through wruff
Ross Wightman committed
074ab13
R
Fix some concerns re timm norms, device factory kwargs, init, pooling/res handling, some formatting
Ross Wightman committed
c43e096
April 18, 2026
Y
Restore ``x.ndim == 4`` dispatch to preserve raw-image + explicit coords path
Yonghye Kwon committed
6045214
Y
Fix gemma4_vit TorchScript failure and exclude from JIT/FX shards
Yonghye Kwon committed
e219c09
April 17, 2026
Y
Align gemma4_vit API with NaFlexVit per PR #2695 review
Yonghye Kwon committed
f152cb5
April 16, 2026
Y
Add Gemma 4 Vision Encoder (gemma4_vit)
Yonghye Kwon committed
a7e09c0
April 15, 2026
R
chore: bump doc-builder SHA for PR upload workflow (#2694)
Remy committed
bc2088e
April 9, 2026
R
Dev version 1.0.27.dev0
Ross Wightman committed
c769a58