b8876
📦 llama-cppView on GitHub →
🐛 1 fixes
Summary
This release includes a fix for speculative decoding by resetting the i_last counter upon low acceptance streaks. It also provides numerous pre-compiled binaries for various operating systems and hardware configurations.
🐛 Bug Fixes
- Reset i_last to zero when a low acceptance streak occurs to ensure the current context is included when rebuilding the speculative map.