llama.cpp b10993 Adds Self-Hosted WebGPU and Vulkan to CI Workflows
llama.cpp release b10993 updates Hugging Face continuous integration (CI) jobs to include self-hosted WebGPU and Vulkan test runs. The release also includes automated binary builds for various platform architectures including Linux, macOS, Windows, and Android. Expanding automated CI testing for WebGPU and Vulkan backends improves code quality and prevents regressions for cross-platform and web-based LLM inference. Automated testing on self-hosted GPU hardware ensures reliable hardware acceleration across non-CUDA ecosystems. The release incorporates pull request #28712, which tested NVIDIA T4 GPU instances, adjusted CPU backend thread configurations, and restored custom execution jobs. Binary releases support CUDA 12/13, ROCm 10.0, SYCL, OpenVINO, and Vulkan backends.
## BACKGROUND
llama.cpp is a popular open-source C/C++ framework enabling high-performance inference of Large Language Models (LLMs) on varied hardware. WebGPU is a modern web API allowing web applications to leverage local graphics hardware for compute-heavy parallel workloads like AI inference.