llama.cpp b9611 improves build compatibility across platforms

llama.cpp, the wildly popular C++ implementation for running large language models locally, just dropped release b9611. Maintained by ggml-org, the project now boasts over 116,000 GitHub stars and 19,500 forks, cementing its status as the de facto standard for on-device LLM inference. This latest release focuses on build infrastructure reliability rather than flashy new features. The headline commit, authored by a project maintainer, tweaks fit.h to avoid unnecessarily including llama-ext.h, a change that streamlines compilation for downstream projects.

Beyond the code fix, b9611 ships with a comprehensive set of pre-compiled binaries. Users on macOS get builds for both Apple Silicon (ARM64 with and without KleidiAI) and Intel x64. Windows users can download CPU-only versions or GPU-accelerated builds with CUDA 12, CUDA 13, and Vulkan backends, plus an on-demand HIP build for AMD GPUs. Linux coverage spans Ubuntu x64/ARM64 with Vulkan, ROCm 7.2, OpenVINO, and SYCL FP32 (disabled). Mobile developers get an Android ARM64 binary and an iOS XCFramework. This wide ecosystem support ensures that whether you're running Mistral on a MacBook, Llama on a gaming PC, or Phi on an Android phone, the latest fixes are readily accessible.