On-policy distillation (OPD) combines student-generated rollouts with dense token-level supervision from a teacher. Existing work has mainly studied its algo…
机构:清华
来源:arXiv 2609.04172 | AI4Papers 论文推荐平台