llama.cpp now builds on 20+ platforms from Apple Silicon to Ascend NPUs

The latest commit (b9285) to the llama.cpp repository introduces a targeted CMake change: the router app is now built exclusively during standalone builds, preventing unwanted compilation when llama.cpp is integrated as a subproject. This small fix is part of a much larger update that officially expands the project’s build matrix to over 20 distinct platform/accelerator combinations.

The supported configurations now span every major OS and architecture. On macOS, both Intel and Apple Silicon (including the KleidiAI-accelerated variant) are covered. Linux users get CPU-only builds for x64, arm64, and s390x, plus Vulkan, ROCm 7.2, OpenVINO, and SYCL (FP32 and FP16). Windows adds CUDA 12 and 13 DLLs, HIP, and SYCL. Mobile support includes iOS XCFramework and Android arm64. Perhaps most notably, the openEuler Linux distribution is now supported with Ascend 310P and 910B NPUs using ACL Graph. This comprehensive support solidifies llama.cpp as the most portable framework for running large language models locally.