Visual on-policy distillation (OPD) improves the training of compact visual autoregressive models by learning from trajectories generated by the current stud…
机构:哈尔滨工业大学(深圳)
来源:arXiv 2608.18183 | AI4Papers 论文推荐平台