28.9M-parameter PLE model, in your browser

A TinyStories language model compiled to WebAssembly. 25M of its parameters sit in a quantized lookup table that can stay on disk.

loading model…

Prompt

backend —

tokens/s —

ms/token —

wasm heap —

memory mode —

disk read/token —

table read —

Model trained on TinyStories — TinyStories: How Small Can Language Models Be and Still Speak Coherent English?, Ronen Eldan and Yuanzhi Li, arXiv:2305.07759.

Settings

Sampling

temperature top-k max tokens seed

Runtime backend keep the 25M table on disk (OPFS) Changing either runtime setting restarts the model.