Change8

b8068

📦 llama-cppView on GitHub →
2 features🐛 1 fixes🔧 2 symbols

Summary

This release introduces SVE optimization for aarch64 in the ggml kernel, improving performance on supported hardware, and includes extensive pre-compiled binaries for multiple platforms.

✨ New Features

  • Implemented SVE in Gemm q4_k 8x8 q8_k Kernel for aarch64 architecture within ggml.
  • Added fallback to NEON code path if SVE 256 is not present to maintain performance.

🐛 Bug Fixes

  • Resolved performance degradation issue when SVE 256 was unavailable by ensuring NEON code is used instead of a generic slow computation.

Affected Symbols