This was part of Frontiers in Online Reinforcement Learning

Miles: Open Source RL for Large MoE Models

Banghua Zhu, University of Washington and NVIDIA