一作ACL 2026Findings· Findings of ACL
Entropy Scheduling in Reinforcement Learning for Large Language Models
Xingjin Wang · Howe Tissue · Lu Wang · Linjing Li · Daniel Dajun Zeng
摘要
We observe that entropy in reinforcement learning functions analogously to the learning rate in LLMs. Maintaining stable entropy, as demonstrated in DAPO, helps stabilize RL training, while rapid entropy annealing (i.e., so-called entropy collapse) accelerates local performance improvement and en...