llama.cpp Release b10224 Adds WebGPU f16 Repeat Support
The llama.cpp project has released version b10224, which introduces support for f16 repeat operations within its WebGPU backend. This update is implemented via pull request #26307 under the ggml-webgpu component. This update enhances the capabilities of the WebGPU backend, allowing for more efficient execution of models that rely on half-precision floating-point (f16) operations directly in web browsers. It represents another step toward enabling high-performance, cross-platform local AI inference without requiring native desktop installations. The release includes pre-built binaries for various platforms including macOS, Linux, Windows, Android, and iOS, supporting backends like CUDA, Vulkan, OpenVINO, and SYCL. However, the macOS Apple Silicon build with KleidiAI enabled remains disabled in this release.
## BACKGROUND
llama.cpp is a highly popular open-source C/C++ library designed for efficient local inference of large language models, serving as the core engine for tools like Ollama. WebGPU is a modern web standard and API that provides web applications with low-level, high-performance access to graphics processing units (GPUs) for hardware acceleration.