作者位次 #3/6ACL 2026Findings· Findings of ACL
STEP : Success-Rate-Aware Trajectory-Efficient Policy Optimization
Yuhan Chen · Yuxuan Liu · Long Zhang · Pengzhi Gao · Jian Luan · WEI LIU
摘要
Multi-turn interaction remains challenging for online reinforcement learning. Current GRPO-based methods—either at the trajectory level or the step level—still suffer from fundamental challenges in multi-turn settings: they allocate sampling uniformly across tasks regardless of difficulty, propag...