Llama.cpp b9610 adds broad GPU backend support including KleidiAI

The llama.cpp project, an open-source C/C++ implementation for running large language models (LLMs) locally, has released version b9610. This release focuses on broadening hardware compatibility by providing pre-built binaries for a wide range of operating systems and GPU backends. For macOS, users on Apple Silicon (arm64) can now optionally enable KleidiAI, an AI acceleration library, while macOS Intel (x64) and iOS XCFramework builds are also included. Linux users benefit from builds for x64 (CPU, Vulkan, ROCm 7.2, OpenVINO, SYCL FP32) and arm64 (CPU, Vulkan), with s390x CPU support added.

On Windows, the release covers x64 and arm64 CPUs, plus GPU backends including CUDA 12.4 and CUDA 13.3 DLLs, Vulkan, and HIP (AMD). Android arm64 CPU builds are also available. Notably, some backends like KleidiAI, SYCL FP32, and OpenEuler builds are listed as DISABLED, indicating they require manual compilation or are experimental. This release continues llama.cpp's mission to make LLM inference accessible on consumer hardware without requiring cloud GPUs, leveraging GGML/GGUF quantized models for memory efficiency.