b9491
📦 llama-cppView on GitHub →
✨ 1 features🐛 2 fixes🔧 1 symbols
Summary
This release focuses on stability by resolving PDL race conditions through modifications to __restrict__ usage in kernel headers, alongside providing numerous platform-specific binary builds.
Migration Steps
- If you were relying on the previous behavior of __restrict__ within PDL kernels, note that it is now disabled by default to prevent race conditions. Performance adjustments based on architecture are now handled internally.
✨ New Features
- Simplified usage of __restrict__ via a new macro.
🐛 Bug Fixes
- Resolved race conditions when using PDL by disabling __restrict__ in PDL kernel headers.
- Added Hopper architecture support to the PDL __restrict__ fix.