On-policy distillation (OPD) can succeed with surprisingly few training questions, motivating a view of questions as carriers of distributional knowledge tra…
机构:UIUC
来源:arXiv 2610.27450 | AI4Papers 论文推荐平台