b9022

ggml-org has tagged version b9022 of llama.cpp, their widely-used C/C++ implementation for running large language models and diffusion models locally. This release is a maintenance-focused update, centered on a significant internal refactor of the diffusion generation example code (PR #22590). The refactoring renames enum values and reorganizes logic to make the codebase more readable and easier to extend. While no new user-facing features are announced, such cleanups are critical for long-term stability and performance optimization, especially as the project supports an ever-growing list of hardware backends.

The release ships prebuilt binaries for a vast array of platforms: macOS on Apple Silicon (both with and without KleidiAI acceleration), Intel macOS, iOS framework, Linux on x86 and ARM (CPU, Vulkan, ROCm 7.2, OpenVINO, SYCL), Android ARM64, and Windows (CPU, CUDA 12, CUDA 13, Vulkan, SYCL, HIP). This breadth underscores llama.cpp's role as the go-to solution for running cutting-edge AI models without cloud dependency. With 108k stars on GitHub, the project continues to be actively maintained by the community under ggml-org, and b9022 represents another step toward robust, cross-platform local AI inference.