The llama.cpp project released build b10472, which includes a fix to skip the Unified Memory Architecture (UMA) override for HIP builds. This change addresses an issue where AMD Accelerated Processing Units (APUs) reported inaccurate memory information via hipMemGetInfo.

  • The update resolves GitHub issue #18159 by preventing MemAvailable from over-promising on systems with small carveouts.
  • macOS Apple Silicon builds with KleidiAI are disabled in this release.
  • Ubuntu ROCm 7.14 support is also disabled.
  • Binaries are provided for Linux, Windows, Android, and macOS across CPU, CUDA, Vulkan, OpenVINO, SYCL, and other backends.