作者位次 #8/10ACL 2026Findings· Findings of ACL
DRA - GRPO : Your GRPO Needs to Know Diverse Reasoning Paths for Mathematical Reasoning
Xiwen Chen · Wenhui Zhu · Peijie Qiu · XUANZHAO DONG · Hao Wang · Haiyu Wu · Huayu Li · Aris Sotiras · Yalin Wang · Abolfazl Razi
摘要
Post-training LLMs with Reinforcement Learning, specifically Group Relative Policy Optimization (GRPO), has emerged as a paradigm for enhancing mathematical reasoning. However, standard GRPO relies on scalar correctness rewards that are often non-injective with respect to semantic content: distin...