Change8

b9717

📦 llama-cppView on GitHub →
1 features🔧 1 symbols

Summary

This release enhances ggml-cpu performance by enabling support for K tails in Power10 MMA Q8/Q4 matmul operations, reducing fallback to mnpack. It also provides numerous pre-compiled binaries for diverse hardware and operating system configurations.

✨ New Features

  • ggml-cpu: Support K tails in Power10 MMA Q8/Q4 matmul, allowing more workloads to utilize the MMA kernel by removing the requirement that K be divisible by kc in the tinyBlas_Q0_PPC tiled matmul path.

Affected Symbols