llama.cpp b10151 boosts Intel GPU builds with parallel SYCL

ggml-org has released llama.cpp version b10151, the latest iteration of the widely-used open-source C++ inference engine for LLaMA-style large language models. The headline feature is the parallelization of SYCL build invocations (commit #25903), which significantly speeds up compilation when targeting Intel GPUs. SYCL is a cross-platform abstraction layer for heterogeneous computing, and this optimization directly benefits developers using Intel GPUs (e.g., integrated or ARC) on Linux and Windows. The release includes prebuilt binaries for Ubuntu x64 (SYCL FP32/FP16), Windows x64 (SYCL), and even OpenEuler variants for Intel Data Center GPU Max series.

Beyond SYCL, llama.cpp b10151 offers an extensive matrix of platform builds: macOS (Apple Silicon with optional KleidiAI, Intel x64, iOS XCFramework), Linux (x64 CPU, arm64, s390x, plus Vulkan, ROCm 7.2, OpenVINO), Windows (CPU, arm64, OpenCL Adreno, CUDA 12 & 13, Vulkan, HIP), and Android arm64. The UI assets bundle is also included for those using the built-in web interface. This breadth makes llama.cpp a go-to choice for running LLMs locally without cloud dependencies — from a Raspberry Pi to a server-grade GPU.