1 paper
Yuhang Yang, Mingxuan Xia, Chao Ye +6
On-policy self-distillation (OPSD) uses a self-teacher conditioned on privileged information to turn the sparse outcome signal of reinforcement learning with verifiable rewards (RL…