# What changed in timm

> 57 changes in timm 1 from its official release notes, upgrade guides and advisories: 2 breaking, 1 security, 19 new APIs and options. Each one cited. Sources checked 2026-10-24. HTML: https://unstale.sh/pypi/timm/

To see which of these apply to a repository, run `npx @unstale/cli init` in it.

## timm 1

### Breaking changes

- Changes input-size selection and fixes test-time preprocessing defaults. (1.0.30) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.30)
- Switches the default position interpolation. (1.0.30) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.30)

### Security

- Pickle loading is hardened to address security concerns. (1.0.28) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.28)

### New APIs and options

- Adds DeepSeek-V4 and V4.1 vision classifiers and encoders with native timm Hub weights. (1.0.30) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.30)
- Adds EfficientViM with native timm Hub weights. (1.0.30) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.30)
- Adds iFormer with native timm Hub weights. (1.0.30) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.30)
- Adds Qwen-Drive-1.0-4B vision tower weights as a qwen3_vit pretrained tag. (1.0.30) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.30)
- Adds Qwen3.5 and Qwen3.8 vision encoders, and the Qwen-Drive-1.0-4B vision tower pretrained tag. (1.0.30) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.30)
- Adds ViT classifier and encoder variants, including classifiers with and without the native spatial merger. (1.0.30) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.30)
- Adds ViT classifier and encoder variants, including classifiers with and without the native spatial merger. (1.0.30) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.30)
- Adds ViT classifier and encoder variants, including classifiers with and without the native spatial merger. (1.0.30) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.30)
- Adds ViTs with EVA and NaFlexViT support and native timm Hub weights. (1.0.30) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.30)
- Adds LingBot-Vision models and pretrained weights. (1.0.29) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.29)
- Adds optional key-only masks for NaFlexVit self-attention. (1.0.29) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.29)
- Adds per-batch image and batch size scheduling for non-NaFlex training, including progressive small-to-large resolution schedules. (1.0.29) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.29)
- Adds PP-LCNetV2 model definitions and pretrained weights, including small, base, and large variants. (1.0.29) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.29)
- Model definitions and pretrained weights are added for EUPE ViT and ConvNeXt models. (1.0.28) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.28)
- Model definitions and pretrained weights are added for EUPE ViT and ConvNeXt models. (1.0.28) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.28)
- NaFlexViT patch-layout support is extended, including support for NaFlex-CLAP. (1.0.28) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.28)
- Adds a flag to unify channels-last patchification support across the data pipeline and NaFlex configuration and models. (1.0.28) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.28)
- TIPSv2 model definitions and pretrained weights are added for DINOv2-style ViTs. (1.0.28) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.28)
- Adds a seed option for reproducible validation. (1.0.28) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.28)

### Performance

- Adds caching for variable-patch interpolation. (1.0.29) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.29)
- Makes position embedding sampling compile-friendly and memory-efficient. (1.0.29) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.29)
- Returns the fused RMSNorm2d result directly. (1.0.29) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.29)

### Behavior and default changes

- Supports NaFlex dictionary inputs. (1.0.28) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.28)

### Fixes and other changes

- Fixes RMS clipping direction. (1.0.30) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.30)
- Fixes learning-rate scaling for convolution weights. (1.0.30) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.30)
- Preserves the caller's gradient mode. (1.0.30) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.30)
- Clears the eval-mode cache on every train() or eval() call to prevent stale attention bias caches. (1.0.30) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.30)
- Fixes mixing buffer shape for non-square and single-band images. (1.0.30) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.30)
- Adds device/dtype factory support for a few models. (1.0.30) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.30)
- Fixes channel metadata for pruned models. (1.0.30) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.30)
- Corrects weight decay when loading checkpoints. (1.0.30) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.30)
- Copies the previous gradient instead of retaining a reference to p.grad. (1.0.30) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.30)
- Fixes models that cannot be created with a meta device context. (1.0.30) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.30)
- Improves parsing, including values containing '=' or non-literals. (1.0.30) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.30)
- Fixes stream reuse and synchronization. (1.0.30) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.30)
- Fixes padding for dilated max pooling. (1.0.30) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.30)
- Fixes factored row normalization for tall matrices. (1.0.29) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.29)
- The b config flag now honors the documented value 0. (1.0.29) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.29)
- Fixes an AttributeError when hparams is None. (1.0.29) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.29)
- The inc config flag now honors the documented value 0. (1.0.29) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.29)
- Classifier resets preserve the head device and dtype. (1.0.29) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.29)
- Fixes device and dtype changes for all classifier resets. (1.0.29) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.29)
- Fixes minmax border sampling and additional mixup, CutMix, and bounding-box range issues. (1.0.29) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.29)
- Fixes the NaFlexVit load_pretrained helper and clarifies its purpose. (1.0.29) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.29)
- Improves model factory path and checkpoint handling. (1.0.29) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.29)
- Fixes a crash when layer_decay is set without an explicit minimum scale. (1.0.29) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.29)
- Fixes positional embedding resampling when grid changes have an equal number of tokens. (1.0.29) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.29)
- Fixes handling of custom input channels for pruned EfficientNet models. (1.0.29) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.29)
- Fixes the foreach path so weight decay is not applied once per device and dtype group. (1.0.29) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.29)
- Fixes ZeroDivisionError when the count is 0. (1.0.28) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.28)
- Inference labels are fixed for models fine-tuned on non-ImageNet datasets. (1.0.28) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.28)
- Fixes handling so reg_token is included in no-weight-decay and stem/embed learning-rate groups. (1.0.28) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.28)
- Support for tensor learning rates is improved. (1.0.28) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.28)
- Optimizer support for torch.compile is improved by using a tensor for the step count. (1.0.28) — [release notes](https://github.com/huggingface/pytorch-image-models/releases/tag/v1.0.28)