GPT-15M Chat

Press "Load model" to run the 15M model in your browser (WebGPU/WASM). ~50 MB download once.

GPT-15M · 12.4M params · trained from scratch on 300M tokens · runs 100% locally