TL;DR: llama.cpp release b9547 introduces a utility fix to skip mmproj downloads alongside standard cross-platform build updates.
Summary: The b9547 release of llama.cpp adds a command line argument fix to skip the mmproj download when a user-supplied mmproj file is already provided. It also distributes standard pre-compiled binaries across multiple platforms including macOS Apple Silicon, iOS, Linux, Android, and Windows CUDA/Vulkan environments. The release disables specific experimental configurations like KleidiAI on macOS arm64 and SYCL FP32 builds on Ubuntu.
Why it matters: It ensures cleaner local developer workflows by preventing redundant model project downloads when custom files are specified. Builders deploying llama.cpp to target devices should verify their platform's specific build compatibility and optimizations for this release.
Source: github.com