b8792
The latest commit fixes CI for Apple Silicon and adds support for Vulkan, CUDA, and OpenVINO backends.
The open-source project llama.cpp, maintained by ggml-org, has pushed a significant new release identified as commit b8792. This update primarily focuses on restoring continuous integration (CI) pipelines for Apple's ecosystem, which had been disabled. The commit re-enables build and test workflows for macOS on both Apple Silicon (arm64) and Intel (x64) architectures, as well as for iOS. This fix is crucial for developers who rely on Macs for building and testing the popular inference library, ensuring compatibility and stability for a key user base.
Beyond the Mac fixes, b8792 expands and refines GPU acceleration support across multiple platforms, a core feature for llama.cpp's performance. The release includes a fix for a Vulkan compilation warning and lists updated support for Vulkan on Ubuntu and Windows. For Windows users, it provides pre-built DLLs for both CUDA 12.4 and CUDA 13.1, offering more flexibility for NVIDIA GPU acceleration. The build matrix also confirms ongoing support for other backends like ROCm, SYCL, HIP, and OpenVINO on Linux, and ACL Graph on openEuler for Huawei Ascend chips. This broadens the hardware options available for running large language models locally with optimized speed.
- Re-enables critical CI workflows for macOS (Apple Silicon/Intel) and iOS development.
- Expands GPU support with Vulkan fixes, CUDA 12.4/13.1 DLLs for Windows, and OpenVINO backend.
- Maintains support for diverse hardware including ROCm, SYCL, HIP, and Ascend AI processors.
Why It Matters
This update unblocks Mac developers and provides more GPU options, making local LLM inference faster and more accessible across different hardware setups.