b9495
📦 llama-cppView on GitHub →
✨ 1 features🐛 1 fixes🔧 3 symbols
Summary
This release focuses on Qwen35 model updates, specifically switching to the post-norm hidden state for MTP and renaming an internal variable. Several platform builds are disabled or updated.
Migration Steps
- The internal name for pre_norm has been renamed to nextn.
✨ New Features
- Updated Qwen35 implementation to use the post-norm hidden state for MTP (Multi-Task Processing).
🐛 Bug Fixes
- Fixed an issue related to step35 in Qwen35.