作者位次 #2/5ACL 2026Findings· Findings of ACL
Entropy-Aware Reshaping of Reinforcement Signals for Multi-Answer Reasoning
Zhi Li · Huidan Xu · Zhen Hu · Yali Du · Ying Liu
摘要
Reinforcement learning with verifiable rewards (RLVR) is a standard post-training paradigm for large language models (LLMs), typically relying on group-wise reward and advantage normalization for stability. In set-valued multi-answer tasks, where multiple outputs may be simultaneously correct, th...