llama.cpp Release b10752 Adds Metal Library Support for XCFrameworks
llama.cpp released build b10752, introducing build support for precompiled Metal libraries (.metallib) within XCFramework binaries on Apple platforms. This enhancement was merged via pull request #28163 to improve Metal GPU acceleration packaging for Apple app developers. This change simplifies developer workflows when embedding llama.cpp into iOS and macOS applications by bundling precompiled GPU shaders directly into XCFramework packages. It enables seamless hardware-accelerated local LLM inference across Apple platforms without requiring manual shader compilation steps at runtime. Release b10752 provides prebuilt binaries for macOS, iOS, Linux, Android, and Windows supporting backends like CUDA 12/13, ROCm, Vulkan, and SYCL. Notable build configurations show that KleidiAI acceleration for macOS arm64 and openEuler builds remain disabled in this release.
## BACKGROUND
Metal is Apple's low-level, hardware-accelerated graphics and compute API, which compiles GPU shaders into binary .metallib files for execution. XCFramework is Apple's binary distribution format that allows developers to package libraries targeting multiple platforms (such as iOS, iOS Simulator, and macOS) into a single bundle. llama.cpp is a popular open-source C/C++ library designed for efficient, local execution of large language models across diverse hardware platforms.