v5.5.4
📦 transformersView on GitHub →
🐛 3 fixes🔧 1 symbols
Summary
This patch release (v5.5.4) primarily addresses several critical bug fixes, focusing on tokenizer regressions and issues related to DeepSpeed ZeRO-3 and Qwen2.5-VL scaling.
🐛 Bug Fixes
- Fixed Kimi-K2.5 tokenizer regression and _patch_mistral_regex Attribute.
- Fixed IndexError with DeepSpeed ZeRO-3 when kernels rotary is active.
- Fixed Qwen2.5-VL temporal RoPE scaling applied to still images.