Reinforcement learning (RL) has become central to post-training of large language models. Recent advances in RL for Mixture-of-Experts (MoE) models have prim…
机构：Microsoft
来源：arXiv 2609.13058 | AI4Papers 论文推荐平台