作者位次 #6/9ACL 2026Findings· Findings of ACL
Revisiting Entropy in Reinforcement Learning for Large Reasoning Models
Renren Jin · Pengzhi Gao · Yuqi Ren · Zhuowen Han · Tongxuan Zhang · Wuwei Huang · WEI LIU · Jian Luan · Deyi Xiong
摘要
Reinforcement learning with verifiable rewards (RLVR) has emerged as a prominent paradigm for enhancing the reasoning capabilities of large language models (LLMs). However, the entropy of LLMs usually collapses during RLVR training, leading to premature convergence to suboptimal local minima and...