Skip to content
Nodus

nodus.examples.rl

View Markdown

Reinforcement learning in one file: teach a small model to spell words backwards.

Run it with python -m nodus.examples.rl. To train on your own task, copy this file and change tasks and reward: a task is a prompt and the answer only reward sees, and reward scores one completion against that answer.

main

Section titled “main”
main() -> None

reward

Section titled “reward”
reward(completion, answer)

1.0 for the exact reversed word, 0.0 for a wrong one, None when the completion holds no answer.