b8744
📦 llama-cppView on GitHub →
✨ 1 features🐛 2 fixes🔧 2 symbols
Summary
This release enables the reasoning budget sampler for gemma4 by updating parameter initialization and fixing a related parsing issue in the thought block handling.
✨ New Features
- Enabled reasoning budget sampler functionality for the gemma4 model by adding necessary start/end tags to common parameters initialization.
🐛 Bug Fixes
- Fixed an issue where the reasoning budget sampler would not activate for gemma4 due to missing thinking tags.
- Modified the gemma4 thought parser to use p.space() instead of p.optional(p.literal("\n")) to correctly handle cases where budget=0 forces the sampler to end before a newline.