On-policy distillation (OPD) improves a student language model by training on its own generated trajectories with supervision from a stronger teacher. Howeve…
机构:CUHK(SZ)
来源:arXiv 2610.21195 | AI4Papers 论文推荐平台