Reinforcement learning (RL) post-training improves the single-sample quality of visual generative models, but deployment often provides additional inference …
机构：字节
来源：arXiv 2611.02366 | AI4Papers 论文推荐平台