← examples · Shakespeare GPT

What light through
yonder window breaks?

A 10 million parameter transformer, trained from scratch in a hand-written C++ ML framework, no PyTorch, on the complete works of Shakespeare. Quantized to int8 so the whole model fits in an 11 MB download.

10.8M parameters
40 epochs trained on a T4
val 1.87 loss
int8 · WASM runs in browser
0.85
150

10M params in the browser aren't fast — expect several seconds for short samples, more for longer ones.

Output ready
Press generate to speak with the Bard.