The llama.cpp project has released build b9984, which includes a change to add Q2_K quantization support to the DMMV reorder path via pull request #25064.
- macOS/iOS: Binaries provided for Apple Silicon (arm64), Intel (x64), and iOS XCFramework; KleidiAI is disabled on Apple Silicon.
- Linux: Builds available for Ubuntu x64, arm64, and s390x CPUs, as well as Vulkan, ROCm 7.2, OpenVINO, and SYCL (FP32/FP16) backends.
- Android: CPU binary provided for arm64.
- Windows: Binaries released for x64 and arm64 CPUs, OpenCL Adreno, CUDA 12.4 and 13.3, Vulkan, OpenVINO, SYCL, and HIP.
- openEuler: Support for x86 (310p, 910b) and aarch64 (310p, 910b) with ACL Graph is included, while the main pull request is marked disabled.
This release provides updated binaries across major operating systems and hardware accelerators, enabling Q2_K quantization in the DMMV reorder path for compatible SYCL builds.