llama.cpp b9198 improves Vulkan builds with SPIRV-Headers check

ggml-org has released llama.cpp version b9198, a maintenance update that refines Vulkan support and expands platform coverage. The key changes include adding a CMake check for SPIRV-Headers, a required dependency for Vulkan projects, and fixing the search path for SPIRV-Headers on macOS Vulkan CI runners. These improvements ensure smoother builds and fewer configuration errors when compiling llama.cpp with Vulkan.

This release ships with over 30 pre-built binaries across major operating systems. macOS users get builds for Apple Silicon (arm64) with and without KleidiAI, Intel x64, and iOS XCFramework. Linux supports CPU-only, Vulkan, ROCm, OpenVINO, and SYCL variants. Windows adds CUDA 12/13, Vulkan, SYCL, and HIP. Android and openEuler are also covered. For developers and AI enthusiasts running local large language models, this means more reliable GPU acceleration via Vulkan and a wider hardware ecosystem to deploy llama.cpp.