llama.cpp Releases Build b10948 with WebGPU Test Fix
llama.cpp has released build b10948, incorporating PR #28855 to exclude the HY_V4 model architecture from WebGPU architecture test runs. The release also supplies updated pre-compiled binaries for a wide range of platforms and execution backends. Excluding non-applicable architectures like HY_V4 from WebGPU integration tests helps prevent false-positive build failures in automated continuous integration pipelines. It ensures that developers using llama.cpp obtain stable pre-built binaries across backends like CUDA, Vulkan, and SYCL. Pull Request #28855, contributed by Stanisław Szymczyk, modifies `test-llama-archs` to skip HY_V4 during WebGPU testing. Pre-compiled binaries released with this build include support for CUDA 12/13, Vulkan, SYCL, ROCm, and OpenVINO across Windows, Linux, macOS, iOS, and Android.
## BACKGROUND
llama.cpp is a popular open-source C/C++ library designed for efficient local LLM inference across diverse hardware backends such as CUDA, Vulkan, WebGPU, and SYCL. To maintain broad platform support, the project continuously issues automated binary builds whenever fixes or architectural adjustments are merged into the main repository.