llama.cpp b10063 adds broad platform support for local LLMs
New release covers macOS, Windows, Linux, Android, and even openEuler with multiple GPU backends.
The ggml-org team released llama.cpp version b10063, a new build of the popular open-source C/C++ inference engine for large language models. The release tag b10063 (commit 7d56da7) includes a sync with the underlying ggml library, ensuring compatibility and performance improvements. Notably, the release boasts an exhaustive matrix of supported platforms and backends: macOS runs on both Apple Silicon (with optional KleidiAI) and Intel x64, plus iOS via XCFramework. Linux builds include Ubuntu for x64, arm64, and s390x CPUs, with Vulkan, ROCm 7.2, OpenVINO, and SYCL (FP32/FP16) variants. Windows users get CPU (x64/arm64) and GPU options including CUDA 12 and 13, Vulkan, OpenVINO, SYCL, and HIP. Android arm64, openEuler (x86 and aarch64 with ACL Graph), and even UI assets are also bundled.
This release reinforces llama.cpp's role as the go-to solution for running LLMs on local hardware without cloud dependency. By supporting such a wide range of architectures—from high-end NVIDIA GPUs (CUDA) to mobile devices (Android, iOS) and niche platforms like openEuler—ggml-org democratizes access to AI inference. Professionals and enthusiasts can now deploy models like Llama, Mistral, or Phi on their own terms, offline, with privacy. The continued expansion of backends (Vulkan for cross-platform GPU, ROCm for AMD, OpenVINO for Intel) means users can leverage existing hardware rather than relying on expensive cloud APIs. While no specific performance metrics were released, the comprehensive platform list suggests optimizations for each target.
- llama.cpp b10063 syncs with ggml and supports macOS (Apple Silicon & Intel), iOS, Linux (multiple backends), Windows (CUDA 12/13, Vulkan, OpenVINO, HIP), Android, and openEuler.
- GPU acceleration options include Vulkan, ROCm, SYCL, and HIP, enabling LLM inference on diverse hardware from AMD to Intel.
- Release includes UI assets and follows 121k stars on GitHub, underscoring its popularity for local AI inference.
Why It Matters
llama.cpp makes private, offline LLM inference viable on personal devices, lowering cost and latency for AI applications.