llama.cpp b9547 adds mmproj download skip, supports all platforms

llama.cpp, the widely-used open-source library for running large language models locally on consumer hardware, has rolled out version b9547. This maintenance release, built automatically via GitHub Actions, introduces a practical fix: when a user supplies their own mmproj (multimodal projection) file, the software now skips downloading it. This addresses issue #24239 and streamlines the setup for users who already have the required projection weights for vision-language models.

The release continues llama.cpp's mission to support diverse hardware. Pre-compiled binaries are available for macOS (Apple Silicon arm64, Intel x64, and even a special KleidiAI-enabled arm64 build), Windows (CPU, CUDA 12/13, Vulkan, SYCL, HIP), Linux (CPU, Vulkan, ROCm 7.2, OpenVINO, SYCL, plus s390x mainframe support), iOS XCFramework, and Android arm64. The extensive build matrix ensures that local LLM inference remains accessible across virtually any setup.