Hugging Face has introduced @huggingface/kernels, a package that contains more than 200 WebGPU kernels. The project is positioned as a resource for local AI, letting developers build browser-based inference workflows that run closer to the user rather than relying on remote servers.
The kernels are built on WebGPU, which gives web applications a standard way to use the GPU. By publishing this set of kernels, Hugging Face is providing building blocks for optimizing AI workloads in the browser and on local devices.
Since this article draws on a single source, there are no competing claims to compare. The announcement reflects Hugging Face’s own description of the package and its intended use.