GPT-22M Chat

Press "Load model" to run the 22M model in your browser (WebGPU/WASM). ~86 MB download once.

GPT-22M · 23.1M params · 24 layers × 256 (GQA) · trained on 600M tokens · runs 100% locally