v4.48.1
📦 transformersView on GitHub →
🐛 3 fixes🔧 3 symbols
Summary
Patch release v4.48.1 fixes a typo in Phi model attention bias, resolves a logic error in gradient accumulation loss, and patches Moonshine's generate wrapper.
🐛 Bug Fixes
- Fixed a regression in Phi models where attention bias was incorrectly set to False.
- Fixed a condition where the Gradient Accumulation (GA) loss bug fix was not being performed.
- Patched Moonshine by removing an issue when wrapping the generate method.
🔧 Affected Symbols
PhiMoonshinegenerate