207 WebGPU kernels put browser AI’s low-level work in the open
A new JavaScript library and versioned kernel collection targeted the operations underneath browser inference.
Article·4 minEXPLORE THE ECOSYSTEM
Inference
Local AI, faster kernels, and getting models into applications.
19 articlesNewest announcements first
Article·4 min
Search can keep more than one vector per document
Late-interaction models preserve finer matching information at the cost of larger indexes.
Article·4 min
Four-bit diffusion is about the compute path, not just smaller weights
Nunchaku integration brings another low-precision option into Diffusers, with hardware compatibility and image quality still central to adoption.
Article·3 min
Reading attention traces before chasing another faster kernel
Attention profiling turns a vague performance problem into a sequence of operations that can be measured and compared.
Article·4 min
Why a shared model implementation can now be a serious serving option
The Transformers backend in vLLM narrows the gap between readable model code and optimized production inference.
Article·4 min
The Kernels update gave low-level compute a dedicated Hub home
A redesigned project added a repository type, revised tools, and broader backend coverage.
Article·3 min
Why browser AI can download the same model twice
An experimental storage proposal explores sharing model resources across web origins.
Article·4 min
Put a local model behind a browser side panel
A Chrome extension guide separates the model runtime from the page interface.
Article·4 min
Build an image pipeline from reusable parts
Modular Diffusers opens up the steps inside a generation pipeline.
Article·4 min
What mixture-of-experts models change for inference
Sparse activation makes parameter counts a less complete guide to runtime behavior.
Article·4 min