The Many Faces of On-Policy Distillation: Pitfalls, Mechanisms, and Fixes 文章

ArXiv CS.AI2026-05-26NEWSen作者: Siqi Zhu, Xuyan Ye, Hongyu Lu, Weiye Shi, Ge Liu
大语言模型
查看原文 →
关系图谱

The Many Faces of On-Policy Distillation: Pitfalls, Mechanisms, and Fixes · 相关技术

相关技术

ORMODEPLALLMlanguage model后训练alignmentTokenStraight-Through EstimatorStanSULReferring expression comprehension (REC)PromptPromiseParts-of-Speech (POS) tagsOn-policy distillationOWLMathematicaMEClarge language modelsHISForEffort Metric AttentionDenseARGANNICE