b8806
📦 llama-cppView on GitHub →
✨ 2 features🐛 2 fixes🔧 2 symbols
Summary
This release introduces the initial Q1_0 backend for CUDA and includes several platform-specific binary builds. Bug fixes include removing unused code and addressing the AMD MMA guard.
✨ New Features
- Initial implementation of the Q1_0 backend for CUDA.
- Attempt to support dp4a (dot product 4-way accumulation).
🐛 Bug Fixes
- Removed unused code.
- Fixed AMD MMA guard.