# What changed in open-clip-torch

> 9 changes in open-clip-torch 3 from its official release notes, upgrade guides and advisories: 7 new APIs and options. Each one cited. Sources checked 2026-10-24. HTML: https://unstale.sh/pypi/open-clip-torch/

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

## open-clip-torch 3

### New APIs and options

- Adds an alternative helper function for text masking. (3.0.0) — [release notes](https://github.com/mlfoundations/open_clip/releases/tag/v3.0.0)
- Adds support for alternative tokenizers for CLIPS. (3.0.0) — [release notes](https://github.com/mlfoundations/open_clip/releases/tag/v3.0.0)
- Adds support for wiring a custom attention block through configuration. (3.0.0) — [release notes](https://github.com/mlfoundations/open_clip/releases/tag/v3.0.0)
- Adds a command-line argument to force the context length. (3.0.0) — [release notes](https://github.com/mlfoundations/open_clip/releases/tag/v3.0.0)
- Adds a schema for loading models and tokenizers from a local folder. (3.0.0) — [release notes](https://github.com/mlfoundations/open_clip/releases/tag/v3.0.0)
- Introduces PE Core. (3.0.0) — [release notes](https://github.com/mlfoundations/open_clip/releases/tag/v3.0.0)
- Adds resizing support for timm image encoders. (3.0.0) — [release notes](https://github.com/mlfoundations/open_clip/releases/tag/v3.0.0)

### Fixes and other changes

- Fixes CustomTextCLIP.no_weight_decay. (3.0.0) — [release notes](https://github.com/mlfoundations/open_clip/releases/tag/v3.0.0)
- Fixes positional embedding interpolation for CustomText models. (3.0.0) — [release notes](https://github.com/mlfoundations/open_clip/releases/tag/v3.0.0)