llama.cpp b11347: Hexagon HTP Skeleton Build Fixes and Broad Platform Binaries Released
The llama.cpp project has released version b11347, fixing issues with Hexagon HTP skeleton catalog dependencies and updating platform binaries. Binaries are available across major operating systems and hardware backends, though openEuler support is currently disabled.
What changed?
llama.cpp release b11347 focuses on updates for Hexagon support, specifically by installing rebuilt Hexagon Tensor Processor (HTP) skeletons and fixing HTP skeleton catalog dependencies. In addition, fresh prebuilt binaries for a wide range of platforms and hardware backends—including macOS (Apple Silicon and Intel), iOS, Ubuntu (multiple GPU/CPU backends), Linux (Snapdragon support), Android (Snapdragon), and Windows—are now available. However, openEuler distribution support is currently disabled.
Why does it matter to an everyday developer?
For developers deploying LLMs on-device or on custom hardware (including Snapdragon NPUs), these Hexagon fixes ensure catalog dependencies are correctly handled, improving stability and correctness of model execution. The extensive availability of prebuilt binaries simplifies initial setup and deployment across many system configurations, reducing friction for cross-platform development and testing. The explicit note that openEuler support is disabled prevents wasted effort for users targeting that platform.
What can the developer do now?
Download the relevant llama.cpp b11347 binaries for your platform—for example, macOS (Apple Silicon or Intel), Ubuntu (CPU, CUDA, Vulkan, ROCm, OpenVINO, SYCL), Windows (CPU, CUDA, Vulkan, etc.), or target Android/Linux Snapdragon backends for on-device AI acceleration. If working with Hexagon NPUs, these improvements should yield a smoother experience. Developers targeting openEuler should wait for future updates that re-enable support.