Reinforcement learning post-training improves the capabilities of large language models, but generating training responses is a major computational bottlenec…
机构：清华
来源：arXiv 2611.06984 | AI4Papers 论文推荐平台