Your Teacher Can't Help You Here: Combating Supervision Fidelity Decay in On-Policy Distillation 文章

ArXiv CS.CL2026-06-01NEWSen作者: Yanjiang Liu, Jie Lou, Xinyan Guan, Yuqiu Ji, Hongyu Lin, Ben He, Xianpei Han, Le Sun, Xing Yu, Yaojie Lu
查看原文 →
关系图谱

Your Teacher Can't Help You Here: Combating Supervision Fidelity Decay in On-Policy Distillation · 相关技术

相关技术

On-policy distillationreverse-KL distillationLookahead Group RewardSupervision Fidelity Decay (SFD)