The llama.cpp project has published the b9929 release, providing binaries across multiple platforms including macOS, Linux, Windows, Android, and openEuler.

  • Vulkan builds reduce the submission threshold based on CU count for small AMD GPUs.
  • macOS Apple Silicon KleidiAI support is disabled in this build.
  • OpenCL Adreno support is added for Windows arm64.
  • ROCm 7.2 binaries are available for Ubuntu x64.

This release updates the library with platform-specific optimizations and availability across various hardware backends.