1 paper · 1 filter
Lianrui Li, Dakuan Lu, Jiawei Shao +1
We introduce Self-correction Relative Policy Optimization (ScRPO), a novel reinforcement learning framework designed to empower large language models with advanced mathematical rea…