Skip to content
Better HN
Beyond 80/20: High-Entropy Minority Tokens Drive Effective RL for LLM Reasoning | Better HN