dapo

Here is 1 public repository matching this topic...

从“为什么能去掉 critic”讲到“训练为什么学不动”：手算组内优势，拆解奖励与归一化，串起 Dr.GRPO、DAPO、GSPO 的改进动机。中文推导、核心代码、训练排坑，含 37 道章末练习与参考答案。

tutorial reinforcement-learning chinese post-training policy-optimization llm gspo grpo rlvr dapo

To associate your repository with the dapo topic, visit your repo's landing page and select "manage topics."