b8858
📦 llama-cppView on GitHub →
✨ 2 features🐛 1 fixes🔧 1 symbols
Summary
This release introduces significant optimizations to the ggml-cpu q1_0 dot product for x86 and generic CPUs, alongside bug fixes related to fp16 conversion. It also provides updated pre-built binaries across numerous platforms.
✨ New Features
- Implemented optimized q1_0 dot for x86 and generic CPU backends.
- Added support for CUDA 13.1 builds on Windows.
🐛 Bug Fixes
- Fixed inconsistency with fp16 conversion for generic q1_0 dot and deduplicated generic fallback.