b10632
📦 llama-cppView on GitHub →
✨ 1 features🔧 2 symbols
Summary
Introduced chunked SSD MMA for Mamba-2 prefill optimization on Metal, improving performance by dropping the scalar SSD path and utilizing MMA with a sequential tail. Several constants and internal states have been refactored for clarity and efficiency.
Migration Steps
- Remove scalar SSD path; MMA + sequential tail is now used.
- Remove state_from_dst and rename CS and NSG constants.
✨ New Features
- Added chunked SSD MMA for Mamba-2 prefill optimization on Metal.