b9689
📦 llama-cppView on GitHub →
✨ 2 features🔧 1 symbols
Summary
This release significantly enhances the Metal backend by adding support for f16 and bf16 data types in the concat operator, along with extending support to i8, i16, and i64.
Migration Steps
- Update device support check to allow f16 unconditionally and bf16 when device supports bfloat16.
✨ New Features
- Metal backend concat operator now supports f16 and bf16 tensor types.
- Metal backend concat operator extended to support i8, i16, and i64 types.