Flowrl Reward Distribution Matching

Train LLMs with distribution-matching rewards instead of reward maximization to achieve 10% improvement on math reasoning while improving solution diversity by matching the full reward distribution via flow balancing, addressing mode collapse in long chain-of-thought tasks.

adu2021 Updated 7 repo stars

File contents

adu2021/skillxiv commit caa7873c9c

Frequently asked questions

npx get-skill add adu2021/flowrl-reward-distribution-matching