The llama.cpp project has released version b10020, which includes a fix for the SYCL backend's get_rows function affecting Q2_K, Q4_K, and Q5_K quantization formats.

  • macOS/iOS builds provided for Apple Silicon (arm64), Intel (x64), and iOS XCFramework, with KleidiAI disabled on Apple Silicon.
  • Linux binaries available for Ubuntu x64, arm64, and s390x CPUs, plus Vulkan, ROCm 7.2, OpenVINO, and SYCL FP32/FP16 variants.
  • Android arm64 CPU build included.
  • Windows releases cover x64 and arm64 CPUs, OpenCL Adreno, CUDA 12.4 and 13.3, Vulkan, OpenVINO, SYCL, and HIP.
  • openEuler support is partially disabled, with builds available for x86 (310p, 910b ACL Graph) and aarch64 (310p, 910b ACL Graph).
  • A standalone UI binary is also provided.

This release ensures broader hardware compatibility across multiple operating systems and accelerators while correcting specific SYCL quantization issues.