Semantic Flow Regularization: Teaching LLMs to Generate Diverse Yet Coherent Responses 文章

ArXiv CS.CL2026-05-28NEWSen作者: Kerui Peng, Feifei Li, Xingyu Fan, Wenhui Que
机器人
查看原文 →
关系图谱

Semantic Flow Regularization: Teaching LLMs to Generate Diverse Yet Coherent Responses · 相关技术

相关技术

cross-entropy objectiveSFTconditional flow matchingSemantic Flow RegularizationCross-Style Collapse