llama.cpp b9839 fixes Tailwind UI, expands hardware support
New release adds CUDA 13, ROCm 7.2, and KleidiAI for Apple Silicon.
The latest release of llama.cpp, tagged b9839, brings a targeted fix for the tools/ui component: restoring Tailwind CSS scanning in ignored worktrees (issue #24879). This patch ensures that UI assets are properly rebuilt when working in Git worktrees that ignore certain files, streamlining development workflows for contributors. Beyond the bug fix, this release continues llama.cpp's tradition of broad hardware support, offering precompiled binaries for nearly every major platform.
Specifically, the release includes builds for macOS (Apple Silicon both with and without KleidiAI acceleration, plus Intel x64 and iOS XCFramework), Linux (x64 and arm64 CPUs, Vulkan, ROCm 7.2, OpenVINO, and SYCL with FP32/FP16), and Windows (x64 and arm64 CPUs, CUDA 12.4 and 13.3 DLLs, Vulkan, OpenCL for Adreno, OpenVINO, SYCL, and HIP). Android arm64 is also covered. This extensive matrix allows developers and researchers to run llama.cpp on everything from a laptop to a server with AMD or NVIDIA GPUs, ensuring efficient local inference with models like Llama 3.1, Mistral, and more.
- Fix: Restored Tailwind CSS scanning in ignored Git worktrees (#24879) for tools/ui.
- New hardware support: macOS KleidiAI option, Linux ROCm 7.2, Windows CUDA 13 DLLs.
- Prebuilt binaries for 20+ platform variants including Android arm64 and iOS XCFramework.
Why It Matters
Llama.cpp keeps local LLM inference accessible across diverse hardware, crucial for privacy and offline AI applications.