GRPO-MA: Multi-Answer Generation in GRPO for Stable and Efficient Chain-of-Thought Training

LRMRe-assign community