**Describe the feature** 来自论文：ProRL: Prolonged Reinforcement Learning Expands Reasoning Boundaries in Large Language Models 看上去效果不错，适合小尺寸模型，里面用到了 GRPO和DAPO的结合，KL+周期性重置的策略 **Paste any useful information** 论文地址：https://arxiv.org/pdf/2505.24864
Describe the feature
来自论文：ProRL: Prolonged Reinforcement Learning Expands Reasoning Boundaries in Large Language Models
看上去效果不错，适合小尺寸模型，里面用到了 GRPO和DAPO的结合，KL+周期性重置的策略
Paste any useful information
论文地址：https://arxiv.org/pdf/2505.24864