b10194
📦 llama-cppView on GitHub →
✨ 1 features🔧 1 symbols
Summary
This release introduces an optimization for ggml-cuda by allowing transpose-free gemmv computation for specific matrix shapes. It also provides pre-compiled binaries for various platforms and hardware accelerators.
✨ New Features
- ggml-cuda: Allow transpose-free gemmv computation for 1xK shaped matrices, leveraging mat_mul_vec_f for potential performance improvements.