On-policy distillation (OPD) trains a student on its own reasoning trajectories using feedback from a stronger teacher. Teacher interventions can improve the…
机构:NTU
来源:arXiv 2609.37510 | AI4Papers 论文推荐平台