小米“天才少女”罗福莉发表新论文，直指AI Agent痛点

\u003cdiv class=\"rich_media_content\"\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e3月16日消息，\u003c/span\u003e\u003c/span\u003e\u003cspan data-pm-slice=\"1 1 [\u0026#34;para\u0026#34;,{\u0026#34;tagName\u0026#34;:\u0026#34;p\u0026#34;,\u0026#34;attributes\u0026#34;:{\u0026#34;style\u0026#34;:\u0026#34;margin-left: 8px; margin-right: 8px;\u0026#34;},\u0026#34;namespaceURI\u0026#34;:\u0026#34;http://www.w3.org/1999/xhtml\u0026#34;}]\" leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e小米AI实验室研究员罗福莉，也就是很多人口中的“天才少女”，又发论文了。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e论文名叫\u003c/span\u003e\u003cspan style=\"font-size: 16px; font-style: italic\" textstyle=\"\"\u003e\u003c!--VERTICAL_CARD_BEGIN_0--\u003eARL-Tangram\u003c!--VERTICAL_CARD_END_0--\u003e: Unleash the Resource Efficiency in Agentic Reinforcement Learning\u003c/span\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e。作者之一，就是罗福莉。\u003c/span\u003e\u003c/span\u003e\u003c!--MID_AD_0--\u003e\u003c!--EOP_0--\u003e\u003c/p\u003e\u003c!--PARAGRAPH_0--\u003e\u003cp\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; text-align: center; white-space: normal\"\u003e\u003c!--IMG_0--\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e如果只看标题，这篇论文似乎只是一个偏工程的研究：\u003c/span\u003e\u003cspan style=\"font-size: 16px; font-weight: bold\" textstyle=\"\"\u003e如何让AI Agent的强化学习训练更省算力\u003c/span\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e但如果把它放在最近几个月AI的发展脉络里，\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e就会发现它其实正好落位在\u003c!--VERTICAL_CARD_BEGIN_1--\u003eOpenClaw\u003c!--VERTICAL_CARD_END_1--\u003e、\u003c!--VERTICAL_CARD_BEGIN_2--\u003eClaude Code\u003c!--VERTICAL_CARD_END_2--\u003e、\u003cspan class=\"\" link-id=\"link-1773642858003-0.4877874011678476\"\u003e\u003c!--VERTICAL_CARD_BEGIN_3--\u003eDevin\u003c!--VERTICAL_CARD_END_3--\u003e\u003c/span\u003e这一波 Agent浪潮的底层。\u003c/span\u003e\u003c/span\u003e\u003c!--MID_AD_1--\u003e\u003c!--EOP_1--\u003e\u003c/p\u003e\u003c!--PARAGRAPH_1--\u003e\u003cp\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; text-align: center; white-space: normal\"\u003e\u003c!--IMG_1--\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e而要理解这篇论文所讲的东西，\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e需要我们先下一个定义，\u003c/span\u003e\u003c/span\u003e\u003cspan data-pm-slice=\"1 1 [\u0026#34;para\u0026#34;,{\u0026#34;tagName\u0026#34;:\u0026#34;p\u0026#34;,\u0026#34;attributes\u0026#34;:{\u0026#34;style\u0026#34;:\u0026#34;margin-left: 8px; margin-right: 8px;\u0026#34;},\u0026#34;namespaceURI\u0026#34;:\u0026#34;http://www.w3.org/1999/xhtml\u0026#34;}]\" leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px; font-weight: bold\" textstyle=\"\"\u003eAI正在从“模型时代”进入“\u003c!--VERTICAL_CARD_BEGIN_4--\u003eAgent时代\u003c!--VERTICAL_CARD_END_4--\u003e”\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e过去几年，大多数 AI 系统的结构其实非常简单。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e输入一段内容，模型计算，然后输出结果。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e从ChatGPT到图像生成模型，本质都是同一个逻辑：\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e输入→模型→输出\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e整个系统的核心资源也很单一，那就是GPU。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e因此过去几年 AI 的竞争基本围绕模型参数规模、训练数据、GPU算力三件事来展开。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e但随着AI Agent的出现，计算结构突然变得复杂起来。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; text-align: center; white-space: normal\"\u003e\u003c!--IMG_2--\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e一个典型的Agent任务可能是这样的：\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e思考→搜索信息→写代码→运行代码→分析结果→再修改\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e在这个过程中，AI不再只调用模型本身，还会不断调用各种工具。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e要用GPU去做模型推理，要用CPU去执行代码，要用API去处理搜索、数据库，可能还要用浏览器进行网页操作......\u003c/span\u003e\u003c/span\u003e\u003c!--MID_AD_2--\u003e\u003c!--EOP_2--\u003e\u003c/p\u003e\u003c!--PARAGRAPH_2--\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e于是，一个Agent系统的资源结构就变成了：\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003eGPU+CPU+API+存储+网络\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e这已经非常接近一个完整的软件系统。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; text-align: center; white-space: normal\"\u003e\u003c!--IMG_3--\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e“帮我整理今天的AI新闻并做成Excel。”\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e一个Agent就会搜索网页、抓取信息、总结内容、生成表格、保存文件，\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e整个过程涉及浏览器、Python、文件系统、模型推理。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e这是Agent计算模式的典型例子，\u003c!--VERTICAL_CARD_BEGIN_5--\u003e龙虾\u003c!--VERTICAL_CARD_END_5--\u003eOpenClaw也是这么做的，\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e而最近\u003cspan class=\"\" link-id=\"link-1773642858128-0.0863631806098788\"\u003eAnthropic\u003c/span\u003e更新的Claude Code，也在逐渐向同样的方向发展。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e/loop：让AI自动循环执行任务\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003eCLAUDE.md：提供长期记忆\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003eSession Spawning：手机远程启动电脑上的AI编程任务\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e连不少开发者看完更新后的第一反应都是“这不就是官方版龙虾吗？”\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; text-align: center; white-space: normal\"\u003e\u003c!--IMG_4--\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e当远程操控、自主执行和长期记忆组合在一起时，AI不再只是一个聊天工具，而更像一个持续运行的\u003c!--VERTICAL_CARD_BEGIN_6--\u003e数字员工\u003c!--VERTICAL_CARD_END_6--\u003e。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e好用，确实是好用，但真正的问题也就在其中，\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px; font-weight: bold\" textstyle=\"\"\u003e算力怎么调度？\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e在传统模型训练中，资源基本只有GPU。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e但在 Agent 系统里，不同任务会不断切换资源类型：\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e有时需要GPU，有时需要CPU，有时需要API\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e如果资源管理方式不合理，就会出现大量浪费。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e论文里给出的典型例子中有一项数据，\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e在某些AI编程任务中，CPU的真实利用率只有47%。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e剩下的算力基本处于闲置状态。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e对于正在进行大规模训练的 AI 公司来说，这种浪费意味着巨大的成本。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e基于此，罗福莉等人的论文提出了一种新的系统：\u003c/span\u003e\u003cspan style=\"font-size: 16px; font-weight: bold\" textstyle=\"\"\u003eTangram\u003c/span\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e它的核心思路很像操作系统的进程调度。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; text-align: center; white-space: normal\"\u003e\u003c!--IMG_5--\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e传统系统的资源分配方式是任务级调度：\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e一个Agent任务开始时，就会占用CPU或GPU，直到任务结束。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e而Tangram采用的是\u003c!--VERTICAL_CARD_BEGIN_7--\u003e动作级调度\u003c!--VERTICAL_CARD_END_7--\u003e。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e也就是说，只有当某个具体操作需要资源时，系统才会分配资源。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e写代码→GPU\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e编译代码→CPU\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e运行测试→CPU\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e搜索资料→API\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e这样不同Agent的任务可以共享资源。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; text-align: center; white-space: normal\"\u003e\u003c!--IMG_6--\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e实验结果显示，这种调度方式带来明显提升：平均任务时间缩短4.3倍，训练效率提升1.5倍，外部资源成本降低71%，\u003c/span\u003e\u003c/span\u003e\u003c!--MID_AD_3--\u003e\u003c!--EOP_3--\u003e\u003c/p\u003e\u003c!--PARAGRAPH_3--\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e对于正在进行大规模Agent训练的团队来说，这种效率提升非常关键。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e如果把最近几个月的技术动向连起来，会看到一条越来越清晰的路线。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003eOpenClaw解决的是Agent的运行框架问题，\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003eClaude Code、Devin解决的是Agent在真实任务中的应用问题，\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e而罗福莉等人做Tangram这样的研究，则开始解决Agent的底层计算问题。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e这说明AI行业正在经历一个结构性变化，\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px; font-weight: bold\" textstyle=\"\"\u003e竞争的焦点正在从模型能力转向Agent系统能力\u003c/span\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e。\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp style=\"caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0); margin-bottom: 25px; margin-left: 8px; margin-right: 8px; white-space: normal\"\u003e\u003cspan leaf=\"\" style=\"color: rgba(0, 0, 0, 0.9); font-size: 17px; letter-spacing: 0.034em; line-height: 1.6\"\u003e\u003cspan style=\"font-size: 16px\" textstyle=\"\"\u003e这也正是为什么OpenClaw、Claude Code、Devin会在今年突然变得如此重要。（转载自AI普瑞斯）\u003c/span\u003e\u003c/span\u003e\u003c!--MID_AD_4--\u003e\u003c!--EOP_4--\u003e\u003c/p\u003e\u003c!--PARAGRAPH_4--\u003e\u003cstyle\u003e.rich_media_content{--news-tabel-th-night-color: #444444;--news-font-day-color: #333;--news-font-night-color: #d9d9d9;--news-bottom-distance: 22px}.rich_media_content p:not([data-exeditor-arbitrary-box=image-box]){letter-spacing:.5px;line-height:30px;margin-bottom:var(--news-bottom-distance);word-wrap:break-word}.rich_media_content .qn-editor-copy p:not([data-exeditor-arbitrary-box=image-box]){letter-spacing:unset;line-height:unset;margin-bottom:unset;word-wrap:unset}.rich_media_content{color:var(--news-font-day-color);font-size:18px}@media(prefers-color-scheme:dark){body:not([data-weui-theme=light]):not([dark-mode-disable=true]) .rich_media_content p:not([data-exeditor-arbitrary-box=image-box]){letter-spacing:.5px;line-height:30px;margin-bottom:var(--news-bottom-distance);word-wrap:break-word}body:not([data-weui-theme=light]):not([dark-mode-disable=true]) .rich_media_content .qn-editor-copy p:not([data-exeditor-arbitrary-box=image-box]):not(.qn-editor-copy){letter-spacing:unset;line-height:unset;margin-bottom:unset;word-wrap:unset}body:not([data-weui-theme=light]):not([dark-mode-disable=true]) .rich_media_content{color:var(--news-font-night-color)}}.data_color_scheme_dark .rich_media_content p:not([data-exeditor-arbitrary-box=image-box]){letter-spacing:.5px;line-height:30px;margin-bottom:var(--news-bottom-distance);word-wrap:break-word}.data_color_scheme_dark .rich_media_content .qn-editor-copy p:not([data-exeditor-arbitrary-box=image-box]){letter-spacing:unset;line-height:unset;margin-bottom:unset;word-wrap:unset}.data_color_scheme_dark .rich_media_content{color:var(--news-font-night-color)}.data_color_scheme_dark .rich_media_content{font-size:18px}.rich_media_content p[data-exeditor-arbitrary-box=image-box]{margin-bottom:11px}.rich_media_content\u003ediv:not(.qnt-video),.rich_media_content\u003esection{margin-bottom:var(--news-bottom-distance)}.rich_media_content hr{margin-bottom:var(--news-bottom-distance)}.rich_media_content .link_list{margin:0;margin-top:20px;min-height:0!important}.rich_media_content blockquote{background:#f9f9f9;border-left:6px solid #ccc;margin:1.5em 10px;padding:.5em 10px}.rich_media_content blockquote p{margin-bottom:0!important}.data_color_scheme_dark .rich_media_content blockquote{background:#323232}@media(prefers-color-scheme:dark){body:not([data-weui-theme=light]):not([dark-mode-disable=true]) .rich_media_content blockquote{background:#323232}}.rich_media_content ol[data-ex-list]{--ol-start: 1;--ol-list-style-type: decimal;list-style-type:none;counter-reset:olCounter calc(var(--ol-start,1) - 1);position:relative}.rich_media_content ol[data-ex-list]\u003eli\u003e:first-child::before{content:counter(olCounter,var(--ol-list-style-type)) '. ';counter-increment:olCounter;font-variant-numeric:tabular-nums;display:inline-block}.rich_media_content ul[data-ex-list]{--ul-list-style-type: circle;list-style-type:none;position:relative}.rich_media_content ul[data-ex-list].nonUnicode-list-style-type\u003eli\u003e:first-child::before{content:var(--ul-list-style-type) ' ';font-variant-numeric:tabular-nums;display:inline-block;transform:scale(0.5)}.rich_media_content ul[data-ex-list].unicode-list-style-type\u003eli\u003e:first-child::before{content:var(--ul-list-style-type) ' ';font-variant-numeric:tabular-nums;display:inline-block;transform:scale(0.8)}.rich_media_content ol:not([data-ex-list]){padding-left:revert}.rich_media_content ul:not([data-ex-list]){padding-left:revert}.rich_media_content table{display:table;border-collapse:collapse;margin-bottom:var(--news-bottom-distance)}.rich_media_content table th,.rich_media_content table td{word-wrap:break-word;border:1px solid #ddd;white-space:nowrap;padding:2px 5px}.rich_media_content table th{font-weight:700;background-color:#f0f0f0;text-align:left}.rich_media_content table p{margin-bottom:0!important}.data_color_scheme_dark .rich_media_content table th{background:var(--news-tabel-th-night-color)}@media(prefers-color-scheme:dark){body:not([data-weui-theme=light]):not([dark-mode-disable=true]) .rich_media_content table th{background:var(--news-tabel-th-night-color)}}.rich_media_content .qqnews_image_desc,.rich_media_content p[type=om-image-desc]{line-height:20px!important;text-align:center!important;font-size:14px!important;color:#666!important}.rich_media_content div[data-exeditor-arbitrary-box=wrap]:not([data-exeditor-arbitrary-box-special-style]){max-width:100%}.rich_media_content .qqnews-content{--wmfont: 0;--wmcolor: transparent;font-size:var(--wmfont);color:var(--wmcolor);line-height:var(--wmfont)!important;margin-bottom:var(--wmfont)!important}.rich_media_content .qqnews_sign_emphasis{background:#f7f7f7}.rich_media_content .qqnews_sign_emphasis ol{word-wrap:break-word;border:none;color:#5c5c5c;line-height:28px;list-style:none;margin:14px 0 6px;padding:16px 15px 4px}.rich_media_content .qqnews_sign_emphasis p{margin-bottom:12px!important}.rich_media_content .qqnews_sign_emphasis ol\u003eli\u003ep{padding-left:30px}.rich_media_content .qqnews_sign_emphasis ol\u003eli{list-style:none}.rich_media_content .qqnews_sign_emphasis ol\u003eli\u003ep:first-child::before{margin-left:-30px;content:counter(olCounter,decimal) ''!important;counter-increment:olCounter!important;font-variant-numeric:tabular-nums!important;background:#37f;border-radius:2px;color:#fff;font-size:15px;font-style:normal;text-align:center;line-height:18px;width:18px;height:18px;margin-right:12px;position:relative;top:-1px}.data_color_scheme_dark .rich_media_content .qqnews_sign_emphasis{background:#262626}.data_color_scheme_dark .rich_media_content .qqnews_sign_emphasis ol\u003eli\u003ep{color:#a9a9a9}@media(prefers-color-scheme:dark){body:not([data-weui-theme=light]):not([dark-mode-disable=true]) .rich_media_content .qqnews_sign_emphasis{background:#262626}body:not([data-weui-theme=light]):not([dark-mode-disable=true]) .rich_media_content .qqnews_sign_emphasis ol\u003eli\u003ep{color:#a9a9a9}}.rich_media_content h1,.rich_media_content h2,.rich_media_content h3,.rich_media_content h4,.rich_media_content h5,.rich_media_content h6{margin-bottom:var(--news-bottom-distance);font-weight:700}.rich_media_content h1{font-size:20px}.rich_media_content h2,.rich_media_content h3{font-size:19px}.rich_media_content h4,.rich_media_content h5,.rich_media_content h6{font-size:18px}.rich_media_content li:empty{display:none}.rich_media_content ul,.rich_media_content ol{margin-bottom:var(--news-bottom-distance)}.rich_media_content div\u003ep:only-child{margin-bottom:0!important}.rich_media_content .cms-cke-widget-title-wrap p{margin-bottom:0!important}\u003c/style\u003e\u003c/div\u003e

【纠错】【责任编辑:你得过脑膜炎】

深度观察

新华全媒头条丨委副总统担任代理总统美对委军事行动引多方谴责