On-policy distillation (OPD) pays twice for each fresh batch: the student generates trajectories and a stronger teacher scores them. Existing methods improve…
机构:阿里
来源:arXiv 2609.24141 | AI4Papers 论文推荐平台