作者位次 #9/12ACL 2026Findings· Findings of ACL
Towards Hierarchical Multi-Step Reward Models for Enhanced Reasoning in Large Language Models
Teng Wang · Jiang Zhangyi · Zhenqi He · Hailei Gong · Shenyang Tong · Wenhan Yang · Zeyu Li · Yanan Zheng · Zifan He · Zewen Ye · Shengjie Ma · Jianping Zhang
摘要
Large Language Models (LLMs) have demonstrated strong mathematical reasoning abilities through supervised fine-tuning and reinforcement learning. However, existing Process Reward Models (PRMs) are vulnerable to reward hacking and require expensive, large-scale annotation of reasoning steps, limit...