Hugging Face releases 207 WebGPU kernels for faster local AI

hugging face has introduced @huggingface/kernels, a minimal library for loading and running optimized webgpu kernels from the hugging face hub. the library comes with an initial collection of 207 kernels, each published as a complete, versioned package. these kernels cover operations used across various machine learning architectures and workloads. the release also includes fleet, an in-browser gpu benchmarking and testing suite. this enables the community to contribute performance and correctness evidence from real-world devices, helping improve kernel variants and optimization decisions. the kernels are apache-2.0 licensed and can be used for faster browser inference.