1 paper
Jiaxuan Wang, Xuan Ouyang, Zhiyu Chen +4
On-policy self-distillation (OPSD) uses a model as its own teacher under privileged context, providing token-level supervision on the model's own reasoning trajectories. However, p…