作者位次 #4/7ACL 2026Findings· Findings of ACL
Efficient PRM Training Data Synthesis via Formal Verification
Ryo Kamoi · Yusen Zhang · Nan Zhang · Sarkar Snigdha Sarathi Das · Ranran Haoran Zhang · Wenpeng Yin · Rui Zhang
摘要
Process Reward Models (PRMs) have emerged as a promising approach for improving LLM reasoning capabilities by providing process supervision over reasoning traces. However, existing approaches for constructing PRM training data remain costly and noisy, as they typically rely on human annotation or...