b8971
📦 llama-cppView on GitHub →
🐛 2 fixes🔧 1 symbols
Summary
This release primarily focuses on bug fixes within ggml-webgpu related to FlashAttention support and kv_tile fitting. It also provides numerous pre-compiled binaries for various operating systems and hardware configurations.
🐛 Bug Fixes
- Fixed FlashAttention support check in ggml-webgpu for devices lacking subgroup support.
- Set path to none if kv_tile does not fit in ggml-webgpu.