llama.cpp b10759 Released with KleidiAI Buffer Initialization Fix
llama.cpp released build b10759, delivering updated pre-built binaries across supported operating systems and platforms. The release includes a specific patch for ggml to avoid initializing KleidiAI buffer types during operation dispatch. Avoiding redundant buffer type initialization on each dispatch reduces overhead and prevents potential execution issues when performing AI matrix operations on Arm CPUs. It ensures smoother performance and better stability for models leveraging Arm's KleidiAI optimizations within llama.cpp. The update incorporates pull request #27891 to fix the KleidiAI buffer initialization behavior in ggml. It also updates automated pre-built binaries for macOS, Windows, Linux, Android, and diverse acceleration backends such as CUDA, Vulkan, ROCm, OpenVINO, and SYCL.
## BACKGROUND
llama.cpp is a popular open-source C/C++ framework designed for high-performance, local inference of Large Language Models, built on top of the ggml machine learning tensor library. Arm KleidiAI is an open-source library providing optimized micro-kernels to accelerate AI workloads directly on Arm architectures.