llama.cpp v0.0.0 b9627 ships with critical UI crash fix and broader platform support

llama.cpp maintainer github‑actions tagged release b9627 on June 13, 2025, addressing a critical bug in the llama‑ui‑embed component. The crash occurred when the UI embed tool was invoked without an explicit asset directory, halting processes that rely on the embedded UI for local LLM interaction. This fix ensures stable operation in configurations where asset paths are not manually specified.

Beyond the bugfix, the release expands the pre‑built binary matrix significantly. For the first time, users on Windows can download builds for CPU, CUDA 12.4, CUDA 13.3, Vulkan, SYCL, and HIP. Linux users gain support for Vulkan on both x64 and arm64, ROCm 7.2, OpenVINO, and SYCL FP16. macOS binaries are available for Apple Silicon (arm64) and Intel (x64), as well as an iOS XCFramework. Android arm64 and Linux s390x are also included. Notably, Apple Silicon with KleidiAI remains disabled in this release. The broad availability of pre‑compiled artefacts removes the need for users to compile llama.cpp from source, dramatically lowering the barrier to running state‑of‑the‑art large language models locally.