Introducing @huggingface/kernels: 200+ WebGPU Kernels for Local AI

Hugging Face has released @huggingface/kernels, a minimal library for loading and running over 200 optimized WebGPU kernels for local, in-browser AI. This initiative aims to accelerate browser inference by providing a foundational layer of high-performance GPU operations. Each kernel is published as a complete, versioned package on the Hub, containing its interface, correctness tests, benchmark cases, and shader templates. Alongside the kernels, a new in-browser benchmarking suite called Fleet has been launched to crowdsource performance and correctness data from a wide range of user hardware.