b8883

The ggml-org team behind the widely-used llama.cpp project has released version b8883, marking a significant infrastructure update for the open-source large language model inference engine. This release focuses on code quality and expanded hardware compatibility, moving all chat conversion functions into a common library with added testing to improve stability and maintainability. The refactoring addresses issue #20690 and represents a cleaner architectural approach that will benefit future development.

The most notable user-facing change is the dramatic expansion of supported hardware backends across all major platforms. For Linux users, new builds now include Vulkan support for both x64 and arm64 architectures, ROCm 7.2 for AMD GPU acceleration, and OpenVINO for Intel hardware optimization. Windows users gain CUDA 12.4 and 13.1 DLLs alongside Vulkan and experimental SYCL/HIP support, while macOS maintains its Apple Silicon and Intel builds with KleidiAI acceleration options. This broad compatibility matrix makes llama.cpp one of the most versatile tools for running LLMs locally on diverse hardware configurations.