Beyond the 80/20 Rule: High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM Reasoning
Paper
•
2506.01939
•
Published
•
187
•
7
Totally Free + Zero Barriers + No Login Required