b8808

The open-source llama.cpp project, maintained by ggml-org, has released commit b8808 as part of its continuous development of efficient local LLM inference. This update primarily addresses server functionality by implementing random media markers (#21962) to improve the reliability of media handling in API responses. The change resolves potential conflicts when serving multiple media files simultaneously, while also cleaning up legacy image token handling and reverting special character processing that caused issues in certain deployments.

The release is notable for its extensive platform support, providing 27 different pre-built binary assets across major operating systems and hardware architectures. For macOS users, there are separate builds for Apple Silicon (both standard and KleidiAI-enabled) and Intel processors, plus iOS XCFrameworks. Linux distributions get comprehensive coverage with CPU builds for x64, arm64, and even s390x architectures, plus accelerated versions for Vulkan, ROCm 7.2, and OpenVINO. Windows users benefit from CUDA 12.4 and 13.1 DLL packages, Vulkan support, and experimental SYCL/HIP builds. The openEuler builds specifically target Huawei's Ascend AI processors (310p and 910b) with ACL Graph optimization, reflecting the project's commitment to diverse hardware ecosystems.