tiny tiny stories

a 7.5M-param ternary language model — … — running entirely in your browser. no server, no gpu.

cold start — first token — — tokens/sec download — loading…
⏱ cold start = page load → first token (incl. model download + init)