Skip to content
#

in-browser-inference

Here are 4 public repositories matching this topic...

A 19.5M parameter next word prediction transformer trained from scratch RMSNorm, RoPE, SwiGLU, grouped query attention quantised to int8 and running entirely in your browser via WebAssembly. Perplexity 28.3, top 5 accuracy 60.7%, 16ms per prediction, zero inference servers. Ships with the writing app, user dashboard, and admin console.

  • Updated Aug 25, 2026
  • TypeScript

Add this topic to your repo

To associate your repository with the in-browser-inference topic, visit your repo's landing page and select "manage topics."

Learn more