Percy Liang

When RL Meets Adaptive Speculative Training: A Unified Training-Serving System

当强化学习遇见自适应推测训练：一个统一的训练-服务系统

Junxiong Wang, Fengxiang Bie, Jisen Li, Zhongzhu Zhou, Zelei Shao, Yubo Wang, Yinghui Liu, Qingyang Wu, Avner May, Sri Yanamandra, Ce Zhang, Tri Dao, Percy Liang, Ben Athiwaratkun, Shuaiwen Leon Song, Chenfeng Xu, Xiaoxia Wu

AI总结 本文提出Aurora系统，通过强化学习实时学习推测器，解决传统方法中部署延迟和领域漂移问题，实验显示在多个模型上实现显著加速。