Tiny AI Lab: Browser-Sized Models on Your Hardware

Field notes

Tiny AI Lab keeps the workflow in the browser: it scans local capabilities, checks runtime loading, downloads selected model weights, and provides chat plus sequential comparison tests. The page lists eight curated models between 135M and 600M, supports WebGPU with WASM/CPU fallback, and measures each selected model through one warm-up followed by seven fixed tests. It also exposes storage controls and export options, while noting that browser memory readings cannot reveal exact free VRAM.