1 paper
Yuhang Zhou, Lizhu Zhang, Yifan Wu +5
On-Policy Distillation (OPD) trains a student model on its own generative trajectories under dense token-level feedback from a stronger teacher, mitigating both the off-policy dist…