年初的 DeepSeek-R1，带来了大模型强化学习（RL）的火爆。无论是数学推理、工具调用，还是多智能体协作，GRPO（Group Relative Policy Optimization）都成了最常见的 RL 算法。GRPO ...

来源：国防部网·中国军号海军舰艇编队搭载中外学员执行远海实习任务A PLAN Task Group Carrying Chinese and Foreign Midshipmen Will Conduct Ocean-going ...

A PLAN Task Group Carrying Chinese and Foreign Midshipmen Will Conduct Ocean-going Training ...