We just open-sourced the world's fastest WebGPU kernels for local AI on Hugging Face

Hugging Face has open-sourced the world's fastest WebGPU kernels for local AI, enabling over 200 common ML operations to run entirely in-browser. These kernels are available on Hugging Face and are being integrated into Transformers.js, ONNX Runtime Web, and LiteRT.js, enhancing local AI capabilities.

This marks the first time WebGPU kernels for local AI have been open-sourced with this level of performance, unlike previous solutions.