browser-llm-fit: check if an AI model fits your browser before downloading weights

init0 · reddit · 2026-09-04

Developer hemanth released browser-llm-fit, an open-source tool tackling common WebGPU crashes when browser-run models exceed maxStorageBufferBindingSize or lack shader-f16 support. It probes client hardware limits and ranks browser-executable models before weights are downloaded. fit('model') tests one model (returning fit and estimated speed, e.g. 45-65 tokens/sec), fit() returns all models sorted by hardware fit. Live demo, GitHub repo, and npm package available.

Original post →

More from Infra

Infra channel →