Developer Tools

llama.cpp v0.0.0 b9907 expands platform support with new builds

New release adds macOS KleidiAI, Windows ARM64, and more platform builds.

Deep Dive

The ggml-org/llama.cpp project has released version b9907, continuing its tradition of broad hardware support for running large language models locally. This release introduces several new build targets including macOS Apple Silicon with KleidiAI acceleration enabled (and a separate non-KleidiAI build), Windows ARM64 CPU support, and Windows ARM64 with OpenCL Adreno GPU acceleration. Also new are Linux s390x CPU builds and Ubuntu x64 with ROCm 7.2 for AMD GPUs. The release drops a few older targets but adds UI assets.

The only code change in b9907 is a missing include fix in common.h (commit c1a411f), which addresses a compilation error on certain platforms. The release is signed with a verified GPG key and brings the total available platforms to more than 20, covering macOS (Intel and Apple Silicon), Linux (x64, ARM64, s390x), Windows (x64, ARM64), Android (ARM64), and various GPU backends (Vulkan, CUDA 12/13, ROCm, OpenVINO, SYCL, HIP, OpenCL). This release underscores llama.cpp's commitment to making local LLM inference accessible across virtually any modern device.

Key Points
  • New builds: macOS ARM64 with KleidiAI, Windows ARM64 CPU, Windows ARM64 OpenCL Adreno
  • Added Linux s390x CPU and Ubuntu x64 with ROCm 7.2 (AMD GPU) builds
  • Single code fix: missing include in common.h (#25220)

Why It Matters

Broader platform coverage means more developers can run local LLMs on diverse hardware, from mainframes to ARM laptops.

📬 Get the top 10 AI stories daily