Reinforcement learning (RL) increasingly targets small language models, and executing these workloads efficiently across multiple accelerators requires effec…
机构：Meta
来源：arXiv 2611.16301 | AI4Papers 论文推荐平台