CJReinforce/PURE

Official code for the paper, "Stop Summation: Min-Form Credit Assignment Is All Process Reward Model Needs for Reasoning"

语言：Python · GitHub 星标：174 · Fork：8

话题：llm · mathematics · o1 · r1 · reasoning · reinforcement-finetuning · reinforcement-learning · rl

在 GitHub 上查看原仓库