llama.cpp b9528 adds UI updates and broader platform support
New release supports CUDA 13, ROCm 7.2, and Android arm64 builds.
The llama.cpp project, maintained by ggml-org, released version b9528 on June 5th. This maintenance update focuses on two areas: UI infrastructure and expanded platform support. The UI component now automatically runs `npm install` when the `package-lock.json` is newer than `node_modules`, ensuring frontend dependencies stay in sync. This is a minor but important quality-of-life fix for users building the web interface.
On the build side, b9528 adds support for the latest GPU toolkits: CUDA 12.4 and CUDA 13.3 on Windows, ROCm 7.2 on Linux, and OpenVINO on Ubuntu. It also continues to support Vulkan, SYCL, HIP, and CPU-only builds across Linux, macOS (Apple Silicon and Intel), Android (arm64), and Windows (x64 and arm64). Notably, some builds (KleidiAI on macOS, SYCL on Windows, and openEuler variants) are disabled in this release. This version ensures that llama.cpp remains compatible with cutting-edge hardware and driver stacks for local LLM inference.
- UI assets update: npm install runs automatically when package-lock.json is newer than node_modules
- New builds: Windows CUDA 12.4 & 13.3, Linux ROCm 7.2, OpenVINO, and Android arm64
- Disabled builds: macOS KleidiAI, Windows SYCL, openEuler (310p & 910b) — possibly temporary
Why It Matters
Keeps llama.cpp compatible with latest GPU toolkits, enabling efficient local LLM inference on diverse hardware.