v0.8.2

teamhanko/hankov0.8.2May 29, 2026by stephantul

AI Summary

Release focusing on training refactoring, multiword token fixes, and skeletoken upgrades.

Key Highlights

Refactor training logic.
Fix multiword token handling and incorrect added tokens.
Upgrade skeletoken dependency.
Add frozen-weights option to training.
Clean up post-processing code.

New Features

Training refactoring
Multiword token fixes
Skeletoken upgrade
Frozen-weights option
Post-processing cleanup

Full Release Notes

## What's Changed
* fix: multiword broken by @stephantul in https://github.com/MinishLab/model2vec/pull/317
* feat: refactor training by @stephantul in https://github.com/MinishLab/model2vec/pull/316
* fix: upgrade skeletoken by @stephantul in https://github.com/MinishLab/model2vec/pull/318
* fix: Incorrect added token can cause issues when adding token as multiword token by @stephantul in https://github.com/MinishLab/model2vec/pull/319
* fix: don't set weights for non-quantized models by @stephantul in https://github.com/MinishLab/model2vec/pull/321
* fix: correctly set weights when w is initialized by @stephantul in https://github.com/MinishLab/model2vec/pull/322
* feat: convert output to float in distill by @stephantul in https://github.com/MinishLab/model2vec/pull/320
* clean up post-processing code by @stephantul in https://github.com/MinishLab/model2vec/pull/323
* chore: fix pre-commit by @stephantul in https://github.com/MinishLab/model2vec/pull/326
* chore: add exclude-newer by @stephantul in https://github.com/MinishLab/model2vec/pull/328
* feat: add frozen-weights option to training by @stephantul in https://github.com/MinishLab/model2vec/pull/327
* fix: small typing issue by @stephantul in https://github.com/MinishLab/model2vec/pull/329
* chore: Update email by @Pringled in https://github.com/MinishLab/model2vec/pull/331
* feat: add trusted publisher by @stephantul in https://github.com/MinishLab/model2vec/pull/332


**Full Changelog**: https://github.com/MinishLab/model2vec/compare/v0.8.1...v0.8.2