[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"site-home-live-entry-setting":3,"site-header-contests":8,"site-header-camps":49,"study-material-detail-study-100":104},{"enabled":4,"title":5,"statusText":5,"buttonText":5,"campSlug":5,"stage":6,"classroomId":7,"hasClassroom":4},false,"",0,null,{"total":9,"list":10,"pageNum":44,"pageSize":45,"size":9,"startRow":44,"endRow":9,"pages":44,"prePage":6,"nextPage":6,"isFirstPage":46,"isLastPage":46,"hasPreviousPage":4,"hasNextPage":4,"navigatePages":47,"navigatepageNums":48,"navigateFirstPage":44,"navigateLastPage":44},3,[11,25,35],{"slug":12,"title":13,"logo":7,"liveUrl":14,"prize":15,"sponsor":16,"description":17,"organizer":18,"startTime":19,"endTime":20,"registrationDeadline":21,"status":22,"dashboardVisibility":23,"createdAt":24},"spring2026","2026春季启元人工智能大赛","https:\u002F\u002Fwww.bilibili.com\u002Fblackboard\u002Flive\u002Flive-activity-player.html?cid=1822703816&quality=0","200,000","麒麟软件有限公司","“2026春季启元人工智能大赛”是由启元实验室主办的面向全国人工智能领域开发者、科研人员及高校学生的权威赛事。大赛旨在促进与培养具备工程实践能力和创新能力的优秀人工智能领域开发者、推广国产软硬件技术与生态、助力国家人工智能产业高质量发展。\n各参赛选手需在规定的比赛周期内，依据赛事规则选定赛题并完成相应挑战，优胜者将共同角逐总额20万元的赛事奖金池，并且还将有机会获得实习和知名企业优质岗位内推资格。","启元实验室","2026-04-19 16:00:00","2026-07-13 16:00:00","2026-06-16 16:00:00","ongoing","private","2026-04-16 08:00:00",{"slug":26,"title":27,"logo":7,"liveUrl":14,"prize":28,"sponsor":7,"description":29,"organizer":18,"startTime":30,"endTime":31,"registrationDeadline":32,"status":33,"dashboardVisibility":23,"createdAt":34},"autumn2025","2025秋季启元人工智能大赛","150,000","“2025秋季启元人工智能大赛”是由启元实验室主办的面向全国人工智能领域开发者、科研人员及高校学生的权威赛事。大赛旨在促进与培养具备工程实践能力和创新能力的优秀人工智能领域开发者、推广国产软硬件技术与生态、助力国家人工智能产业高质量发展。","2025-11-08 08:00:00","2026-01-13 08:00:00","2025-12-30 08:00:00","ended","2025-11-03 08:00:00",{"slug":36,"title":37,"logo":7,"liveUrl":14,"prize":38,"sponsor":7,"description":39,"organizer":18,"startTime":40,"endTime":41,"registrationDeadline":42,"status":33,"dashboardVisibility":23,"createdAt":43},"summer2025","启元人工智能大赛(系统方向)","80,000","启元人工智能大赛由启元实验室主办，其中系统赛道基于启元实验室九源软件栈的相关开源项目，通过拉动开源社区力量，一方面促进国产智能计算生态的发展，另一方面也为人工智能系统方向的学生及从业者提供一个实践平台。","2025-07-08 08:00:00","2025-08-25 08:00:00","2025-07-22 08:00:00","2025-06-22 08:00:00",1,5,true,8,[44],{"total":45,"list":50,"pageNum":44,"pageSize":45,"size":45,"startRow":44,"endRow":45,"pages":44,"prePage":6,"nextPage":6,"isFirstPage":46,"isLastPage":46,"hasPreviousPage":4,"hasNextPage":4,"navigatePages":47,"navigatepageNums":103,"navigateFirstPage":44,"navigateLastPage":44},[51,62,73,84,94],{"id":52,"slug":53,"title":54,"logo":55,"publicLiveEmbedUrl":14,"description":56,"introduction":57,"organizer":18,"coOrganizer":5,"sponsor":16,"startTime":58,"endTime":59,"status":22,"dashboardVisibility":60,"createdAt":58,"updatedAt":61},"019eeeff-6b00-7d55-a629-efafded3e232","summer2026","2026年夏季InfiniTensor大模型与人工智能系统训练营","https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fuploads\u002F2026-06\u002F019eef7f-7816-7828-9309-79c2c61dd568.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=VJSZCyae51iopwoi0izQIKD64uI%3D","由启元实验室发起的InfiniTensor大模型与人工智能系统训练营，面向开源社区、高校及产业界伙伴开放。活动以「普及AI全栈知识技术、培养工程实践与创新型人才、推广国产软硬件技术生态」为核心目标，助力国家人工智能产业高质量发展。","...","2026-06-22 08:00:00","2026-10-19 08:00:00","public","2026-10-12 08:00:00",{"id":63,"slug":64,"title":65,"logo":66,"publicLiveEmbedUrl":7,"description":67,"introduction":57,"organizer":18,"coOrganizer":5,"sponsor":68,"startTime":69,"endTime":70,"status":33,"dashboardVisibility":23,"createdAt":71,"updatedAt":72},"019af3f3-e06a-7245-aa8d-37dc99a69c73","winter2025","2025年冬季InfiniTensor大模型与人工智能系统训练营","https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fuploads\u002F2025-12\u002F019afc69-b7ca-7805-91fb-605429671c01.png?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=wHs6e9DQam6eCjdRCa17%2BAxk2V8%3D","由启元实验室发起的InfiniTensor大模型与人工智能系统训练营，面向开源社区、高校及产业界伙伴开放。活动以「普及AI全栈知识技术、培养工程实践与创新型人才、推广国产软硬件技术生态」为核心目标，助力国家人工智能产业高质量发展。 \n","新华三技术有限公司","2025-12-25 08:00:00","2026-03-27 08:00:00","2025-12-06 08:00:00","2026-10-22 20:24:30",{"id":74,"slug":36,"title":75,"logo":76,"publicLiveEmbedUrl":7,"description":77,"introduction":57,"organizer":18,"coOrganizer":5,"sponsor":78,"startTime":79,"endTime":80,"status":81,"dashboardVisibility":23,"createdAt":82,"updatedAt":83},"01980342-208c-7803-b3f3-19e1d31960ba","2025年夏季InfiniTensor大模型与人工智能系统训练营","https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fuploads\u002F2025-07\u002F0198045e-28fe-78f0-95fe-23835ce58395.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=JKqlkIO2CSuyOH0jhDR9YJP0gYU%3D","InfiniTensor大模型与人工智能系统训练营由启元实验室及清华大学 AI 系统方向团队发起，面向开源社区、高校和产业界，普及人工智能各方面知识和技术，培养相关人才。训练营联合开源项目开发者、学术界大牛、工业界资深工程师参与授课与项目指导。课程采用「线上直播授课 + 在线答疑 + 项目实践辅导 」的方式进行，确保每位同学都能获得充分成长。训练营全程免费，通过视频、直播课、习题和项目实践，在人工智能飞速发展的背景下，为学员提供一窥深层技术的机会。","北京积算科技有限公司","2025-07-14 08:00:00","2025-09-30 08:00:00","upcoming","2025-07-13 08:00:00","2026-10-22 20:24:58",{"id":85,"slug":86,"title":87,"logo":88,"publicLiveEmbedUrl":7,"description":89,"introduction":57,"organizer":18,"coOrganizer":5,"sponsor":5,"startTime":90,"endTime":91,"status":81,"dashboardVisibility":23,"createdAt":92,"updatedAt":93},"01980cf9-a7d8-7796-ae86-0bbe05334add","winter2024","2024年冬季InfiniTensor大模型与人工智能系统训练营","https:\u002F\u002Ftesting-cdn.infinitensor.com\u002Fuploads\u002F2025-07\u002F01981269-ce23-7624-a287-9f07dd0d85f6.jpg","InfiniTensor大模型与人工智能系统训练营由清华大学发起，面向开源社区、高校和产业界，普及人工智能各方面知识和技术，培养相关人才。训练营全程免费，通过视频、直播课、习题和项目实践，在人工智能飞速发展的背景下，为学员提供一窥深层技术的机会。报名学员请阅读“导学”课程页的训练营介绍。","2025-01-05 08:00:00","2025-03-22 08:00:00","2025-07-15 08:00:00","2026-10-22 20:24:39",{"id":95,"slug":96,"title":97,"logo":98,"publicLiveEmbedUrl":7,"description":99,"introduction":57,"organizer":18,"coOrganizer":5,"sponsor":5,"startTime":100,"endTime":101,"status":81,"dashboardVisibility":23,"createdAt":92,"updatedAt":102},"01980c20-363a-7dc3-8f5a-58a68e08dcc4","summer2024","2024年夏季InfiniTensor大模型与人工智能系统训练营","https:\u002F\u002Ftesting-cdn.infinitensor.com\u002Fuploads\u002F2025-07\u002F01981269-fbfa-788a-afcd-17fa9be74ff9.jpg","InfiniTensor大模型与人工智能系统训练营由清华大学发起，面向开源社区和高校，普及人工智能各方面知识和技术，培养相关人才。训练营全程免费，通过课程、习题和项目实践，在人工智能飞速发展的背景下，为学员提供一窥深层技术的机会。","2024-07-20 08:00:00","2024-09-29 08:00:00","2026-10-22 20:24:49",[44],{"currentItem":105,"processedData":125},{"img":106,"keywords":107,"treeType":44,"description":108,"showStatus":44,"updateTime":109,"type":110,"title":111,"content":112,"nextArticle":113,"createTime":109,"sortOrder":117,"canonicalPath":118,"id":119,"contentType":120,"prevArticle":121},"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_fm.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=3vgJ45n2FpltDoytA9RdXZI1cJk%3D","银河麒麟,智算操作系统,Agent 运行时,国产异构,KylinDLA","2026 夏季 InfiniTensor 训练营《麒麟 Agent 运行时系统和智算软件栈优化方法》完整讲解银河麒麟全系列国产操作系统产品；拆解统一智算软件栈异构底座；介绍麒麟 Agent 运行时模型调度、权限安全、多会话内存管理核心能力，配套数字员工、教学一体机等私有化行业落地方案。","2026-10-22 20:38:11",2,"2026 夏季 InfiniTensor 训练营 AI 编程与智能体应用 9：《麒麟 Agent 运行时系统和智算软件栈优化方法》","![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw01.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=h1s7OhjvHYrQXxpqblWvHuC%2BbLw%3D)\r\n\r\n\u003Cspan style=\"font-size:12px;color:#555;\">8 月 11 日，2026 夏季 InfiniTensor 训练营 AI 编程与智能体应用第九课《麒麟 Agent 运行时系统和智算软件栈优化方法》开讲。\r\n上一课《大模型自主智能体》从理论层面讲解大模型自主智能体技术链路，包含 ReAct 循环、三类工具学习范式、X-Agent、ChatDev 多角色协作等通用框架相关内容。\u003C\u002Fspan>\r\n\r\n本节课从操作系统视角解析 Agent 运行时与 OS 的关系及软件栈优化方法，介绍麒麟软件产品体系，讲解智算操作系统如何屏蔽异构算力差异、AI 软件栈等内容。\r\n\r\n\r\n## 麒麟软件与国产操作系统产品体系\r\n### 麒麟发展历程\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw02.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=btGWrWdmrTIXiKYZnavYxd31KrY%3D)\r\n\r\n1. 源头：银河巨型机时代自研系统；863 专项孵化银河麒麟；\r\n2. 分支：中标麒麟面向通用桌面、服务器场景；\r\n3. 整合：2020 年银河麒麟与中标麒麟合并为麒麟软件，多次斩获国家科技进步一等奖。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw03.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=REquScWc2eLUltsV7SF9260A5YU%3D)\r\n\r\n### 核心操作系统产品\r\n1. **OpenKylin 开源社区**：商业版上游开源基座，内置 wlcom 显示合成器、UKUI 桌面。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw04.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=clCp4B3KY7Mpe8kVXa1I1WDEj%2BU%3D)\r\n\r\n2. **银河麒麟桌面 V11**：磐石架构、KSAF 可编程安全框架、原生 AI PC 端侧 AI 适配。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw05.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=qK768zd6Lj9a3qVmYefWZJHVq%2FI%3D)\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw06.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=ZuOc3Mjr3qA%2BuV%2BZ9jiwTehNFNA%3D)\r\n\r\n3. **银河麒麟服务器 V11**：深度融合 AI 能力，支持虚拟化、资源调度等服务器场景。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw07.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=hJQJGTXqES9FFAeMVqJejNDCsZg%3D)\r\n\r\n4. **星光麒麟万物智联系统**：面向万物智联场景的智能终端操作系统，采用统一应用框架，支持多种国产移动 SoC。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw08.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=dyfhhde7L%2BFH0MSWmOFWLthH%2BAo%3D)\r\n\r\n5. **银河麒麟工业操作系统**：轻量桌面、实时隔离、原子更新，适配产线嵌入式设备。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw09.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=pKo93FrIX%2FgKGz16luALXYRqv0I%3D)\r\n\r\n6. **银河麒麟智算操作系统**：面向人工智能大模型训练和推理场景。支持异构混训\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw10.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=mnNPnwrTkFPyQ2LhhEBUrsxuOqk%3D)\r\n\r\n### 生态适配规模\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw11.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=ER2egxXDBVqzoF%2BH2JLgD2454yo%3D)\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw12.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Rn%2FtKZqbyjx344uAmn6WLvsIbkw%3D)\r\n\r\n\r\n## 智算操作系统和智算软件栈\r\n### 国家政策环境\r\n国家 AI+ 行动、政务大模型适配指引要求算力一地建设、多地复用，需操作系统提供统一异构调度、数据安全隔离能力。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw13.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=qva0ggTKbCPxDvjvQI1klap2j14%3D)\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw14.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=CCtfv0UhhJm7Pwi2wJFCLBJyHjY%3D)\r\n\r\n### 国产 AI 软件栈发展现状\r\n1. 国产异构芯片算力存在差距；\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw15.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=1vIPaykXYJ7qKFZWSRPJGGziEgg%3D)\r\n\r\n2. 各厂商编程框架、推理引擎、算子库、通信接口不统一，应用迁移成本极高；\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw16.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=t%2FxEx0RbMmKyh6YhRiqkUkzdQ7Y%3D)\r\n\r\n3. 多卡跨节点通信无标准化方案，混合训练推理吞吐损失严重。\r\n\r\n### 智算操作系统核心目标\r\n构建屏蔽部分硬件差异的统一 AI 软件底座，降低模型在不同芯片平台上的适配与部署成本。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw17.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=qbrWwmOchXC7JcSh62uv8EavAgY%3D)\r\n\r\n\r\n#### 1.银河麒麟深度学习框架 KylinDLA\r\n- 统一适配多版本 PyTorch，兼容英伟达、AMD、国产各类异构加速卡，抹平框架版本差异。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw18.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=5ji9VKzvKL0417kKezFrBios5ZY%3D)\r\n\r\n#### 2.面向国产 CPU+GPU 的算子优化\r\n- 算子融合：合并连续计算，削减 HBM 访存读写；\r\n- 内存调度：KV Cache 分页、张量复用，降低峰值显存占用；\r\n- CPU\u002FGPU 协同分工，均衡负载；\r\n- 已完成多个国产 GPU 专用算子深度调优。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw19.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=V5sXT6NZzEgr3ezozb7y8G8SA3M%3D)\r\n\r\n#### 3. 异构加速卡统一通信库\r\n分层模块化设计，分为策略层、模型层、机制层：\r\n- 同芯片：调用厂商原生通信接口；\r\n- 跨异构芯片： 通过 RDMA 等高速互联技术实现不同计算设备间的数据传输与通信优化；\r\n- 内置 AllReduce \u002F AllGather 集合通信优化，支持多卡混合训练、推理。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw20.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=NtyWQzHNMzHfu%2BMHIb3tEKoOYCw%3D)\r\n\r\n#### 4. 麒麟智算操作系统\r\n- 独立开发沐曦、天枢、海光等芯片专属算子库，支持 FP 8 \u002F INT 8 量化、GQA \u002F MQA 注意力、矩阵乘，7B \u002F 72 B 大模型推理性能较原厂提升显著。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw21.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Cx9vQeq6yOPiV5FKNrG0t5%2FNSyY%3D)\r\n\r\n### 上层模型生态适配\r\n- 已适配大语言、多模态视觉、语音、推荐等200+ 主流模型，配套智算生态网站一键部署。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw22.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=labxNqg5elbwUHbqT1FOlR%2BZH58%3D)\r\n\r\n### 基于全国产智算底座的一体机解决方案\r\n\r\n1. **数字员工一体机**：问答、文档处理、翻译、会议多场景封装，私有化部署；\r\n2. **AI 教学一体机**：配套训练推理实操题目，适配训练营课程开发调试；\r\n3. **编程助手一体机**：内网代码生成、工作流自动化、代码评审；\r\n4. **智慧客服一体机**：专属知识库、多轮对话、工单自动生成。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw23.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=HiFjaCxy31XGRO13iXlXgYFCS7E%3D)\r\n\r\n## 端侧 AI 软件栈和 Agent 运行时系统\r\n\r\n### 端侧 AI 架构\r\n端侧硬件约束：算力有限、带宽低、显存小，侧重功耗与算力平衡。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw24.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=I1vcKUFgYAVzWa0e%2F6%2FhnTXJW2A%3D)\r\n\r\n### 端侧 MoE 模型推理优化\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw25.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=9shS08qzFuiIv1WazfQ7%2BEccboA%3D)\r\n\r\n#### 1. 降低路由 TopK：减少专家激活数量，降低计算量；\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw26.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=71NJqJh7wEFtp8t5BA815cMlBks%3D)\r\n\r\n#### 2. 缩小单个专家：\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw27.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=avI0ATH%2BMXjMvop5c35oHjidARU%3D)\r\n\r\n#### 3. 路由算法优化：简单输入少激活专家，复杂输入多专家协同；\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw28.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=uPUa1WrAXNCNJGVcpy9mvUjfoGg%3D)\r\n\r\n#### 4. 推理部署优化：专家卸载；推测解码。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw29.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=vmUIceDien8GBQib9Hg1ha%2BjlRI%3D)\r\n\r\n### 推理性能 Profiler 分析工具\r\n算子级追踪 Kernel 耗时、访存开销，自动给出剪枝、量化、算子融合优化建议，适配全系列国产 NPU \u002F GPU。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw30.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=CiC43wVxaZ%2FNVOes3Dnp3c3R7S8%3D)\r\n\r\n### 注意力算子优化（面向 GQA \u002F MQA）\r\n重构内存排布，合并单 Block 内多头计算，提升 Warp 归约效率，访存下降，整体性能提升 20% +。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw31.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=fc9ZA9NctSrKRaKIZN5wirLRz58%3D)\r\n\r\n### Sampler 采样算子融合\r\n整合 TopK \u002F TopP \u002F Temperature 全套逻辑，消除多次 Load \u002F Store 访存冗余，Batch 越大加速效果越明显。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw32.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=xgULZwHUE%2B5QhZyu4NyNpasgkF4%3D)\r\n\r\n### 量化访存优化（GPTQ、AWQ）\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw33.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=nMKDz4L5HWb9Vq%2FmYsGhCRbSAnI%3D)\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw34.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=HNeZz%2FdogSl%2BQXhf6ZGfz78Qd%2F0%3D)\r\n\r\n\r\n\r\n\r\n## 加入训练营\r\n📺**完整课程内容，请观看直播或查看课程回放：**\r\n\r\n**InfiniTensor 官网**： https:\u002F\u002Fwww.infinitensor.com\r\n\r\n**B站直播**： InfiniTensor 官方直播间\r\n\r\n**视频号直播**： InfiniTensor 视频号预约\r\n\r\n**答疑交流**： 训练营官方社群⬇️\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002Fqun_1072.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=dUe%2BM0SLl6KCrp97rPdQhphT%2Bck%3D)\r\n\r\n## 关注与交流\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002Fportal_news\u002Fnews_ewm02.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=6FVLIDG%2FuTYOS25cEBcGpul%2Bzno%3D)",{"publishTime":7,"createTime":109,"canonicalPath":114,"id":115,"title":116},"\u002Fstudy-material\u002Fstudy-99","study-99","2026 夏季 InfiniTensor 训练营训练系统 5：《强化学习基础与 DeepSeek 训练介绍》",10,"\u002Fstudy-material\u002Fstudy-100","study-100","markdown",{"publishTime":7,"createTime":109,"canonicalPath":122,"id":123,"title":124},"\u002Fstudy-material\u002Fstudy-101","study-101","2026 夏季 InfiniTensor 训练营推理引擎 6：《InfiniLM》",{"htmlContent":126,"headings":127},"\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw01.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=h1s7OhjvHYrQXxpqblWvHuC%2BbLw%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>\u003Cspan style=\"font-size:12px;color:#555;\">8 月 11 日，2026 夏季 InfiniTensor 训练营 AI 编程与智能体应用第九课《麒麟 Agent 运行时系统和智算软件栈优化方法》开讲。\u003Cbr>上一课《大模型自主智能体》从理论层面讲解大模型自主智能体技术链路，包含 ReAct 循环、三类工具学习范式、X-Agent、ChatDev 多角色协作等通用框架相关内容。\u003C\u002Fspan>\u003C\u002Fp>\n\u003Cp>本节课从操作系统视角解析 Agent 运行时与 OS 的关系及软件栈优化方法，介绍麒麟软件产品体系，讲解智算操作系统如何屏蔽异构算力差异、AI 软件栈等内容。\u003C\u002Fp>\n\u003Ch2 id=\"麒麟软件与国产操作系统产品体系\">\u003Ca class=\"anchor\" href=\"#麒麟软件与国产操作系统产品体系\" aria-label=\"跳转到标题：麒麟软件与国产操作系统产品体系\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>麒麟软件与国产操作系统产品体系\u003C\u002Fh2>\u003Ch3 id=\"麒麟发展历程\">\u003Ca class=\"anchor\" href=\"#麒麟发展历程\" aria-label=\"跳转到标题：麒麟发展历程\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>麒麟发展历程\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw02.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=btGWrWdmrTIXiKYZnavYxd31KrY%3D\" alt=\"\">\u003C\u002Fp>\n\u003Col>\n\u003Cli>源头：银河巨型机时代自研系统；863 专项孵化银河麒麟；\u003C\u002Fli>\n\u003Cli>分支：中标麒麟面向通用桌面、服务器场景；\u003C\u002Fli>\n\u003Cli>整合：2020 年银河麒麟与中标麒麟合并为麒麟软件，多次斩获国家科技进步一等奖。\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw03.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=REquScWc2eLUltsV7SF9260A5YU%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"核心操作系统产品\">\u003Ca class=\"anchor\" href=\"#核心操作系统产品\" aria-label=\"跳转到标题：核心操作系统产品\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>核心操作系统产品\u003C\u002Fh3>\u003Col>\n\u003Cli>\u003Cstrong>OpenKylin 开源社区\u003C\u002Fstrong>：商业版上游开源基座，内置 wlcom 显示合成器、UKUI 桌面。\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw04.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=clCp4B3KY7Mpe8kVXa1I1WDEj%2BU%3D\" alt=\"\">\u003C\u002Fp>\n\u003Col start=\"2\">\n\u003Cli>\u003Cstrong>银河麒麟桌面 V11\u003C\u002Fstrong>：磐石架构、KSAF 可编程安全框架、原生 AI PC 端侧 AI 适配。\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw05.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=qK768zd6Lj9a3qVmYefWZJHVq%2FI%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw06.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=ZuOc3Mjr3qA%2BuV%2BZ9jiwTehNFNA%3D\" alt=\"\">\u003C\u002Fp>\n\u003Col start=\"3\">\n\u003Cli>\u003Cstrong>银河麒麟服务器 V11\u003C\u002Fstrong>：深度融合 AI 能力，支持虚拟化、资源调度等服务器场景。\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw07.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=hJQJGTXqES9FFAeMVqJejNDCsZg%3D\" alt=\"\">\u003C\u002Fp>\n\u003Col start=\"4\">\n\u003Cli>\u003Cstrong>星光麒麟万物智联系统\u003C\u002Fstrong>：面向万物智联场景的智能终端操作系统，采用统一应用框架，支持多种国产移动 SoC。\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw08.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=dyfhhde7L%2BFH0MSWmOFWLthH%2BAo%3D\" alt=\"\">\u003C\u002Fp>\n\u003Col start=\"5\">\n\u003Cli>\u003Cstrong>银河麒麟工业操作系统\u003C\u002Fstrong>：轻量桌面、实时隔离、原子更新，适配产线嵌入式设备。\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw09.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=pKo93FrIX%2FgKGz16luALXYRqv0I%3D\" alt=\"\">\u003C\u002Fp>\n\u003Col start=\"6\">\n\u003Cli>\u003Cstrong>银河麒麟智算操作系统\u003C\u002Fstrong>：面向人工智能大模型训练和推理场景。支持异构混训\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw10.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=mnNPnwrTkFPyQ2LhhEBUrsxuOqk%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"生态适配规模\">\u003Ca class=\"anchor\" href=\"#生态适配规模\" aria-label=\"跳转到标题：生态适配规模\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>生态适配规模\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw11.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=ER2egxXDBVqzoF%2BH2JLgD2454yo%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw12.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Rn%2FtKZqbyjx344uAmn6WLvsIbkw%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch2 id=\"智算操作系统和智算软件栈\">\u003Ca class=\"anchor\" href=\"#智算操作系统和智算软件栈\" aria-label=\"跳转到标题：智算操作系统和智算软件栈\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>智算操作系统和智算软件栈\u003C\u002Fh2>\u003Ch3 id=\"国家政策环境\">\u003Ca class=\"anchor\" href=\"#国家政策环境\" aria-label=\"跳转到标题：国家政策环境\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>国家政策环境\u003C\u002Fh3>\u003Cp>国家 AI+ 行动、政务大模型适配指引要求算力一地建设、多地复用，需操作系统提供统一异构调度、数据安全隔离能力。\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw13.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=qva0ggTKbCPxDvjvQI1klap2j14%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw14.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=CCtfv0UhhJm7Pwi2wJFCLBJyHjY%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"国产-ai-软件栈发展现状\">\u003Ca class=\"anchor\" href=\"#国产-ai-软件栈发展现状\" aria-label=\"跳转到标题：国产 AI 软件栈发展现状\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>国产 AI 软件栈发展现状\u003C\u002Fh3>\u003Col>\n\u003Cli>国产异构芯片算力存在差距；\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw15.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=1vIPaykXYJ7qKFZWSRPJGGziEgg%3D\" alt=\"\">\u003C\u002Fp>\n\u003Col start=\"2\">\n\u003Cli>各厂商编程框架、推理引擎、算子库、通信接口不统一，应用迁移成本极高；\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw16.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=t%2FxEx0RbMmKyh6YhRiqkUkzdQ7Y%3D\" alt=\"\">\u003C\u002Fp>\n\u003Col start=\"3\">\n\u003Cli>多卡跨节点通信无标准化方案，混合训练推理吞吐损失严重。\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Ch3 id=\"智算操作系统核心目标\">\u003Ca class=\"anchor\" href=\"#智算操作系统核心目标\" aria-label=\"跳转到标题：智算操作系统核心目标\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>智算操作系统核心目标\u003C\u002Fh3>\u003Cp>构建屏蔽部分硬件差异的统一 AI 软件底座，降低模型在不同芯片平台上的适配与部署成本。\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw17.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=qbrWwmOchXC7JcSh62uv8EavAgY%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"1-银河麒麟深度学习框架-kylindla\">\u003Ca class=\"anchor\" href=\"#1-银河麒麟深度学习框架-kylindla\" aria-label=\"跳转到标题：1.银河麒麟深度学习框架 KylinDLA\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>1.银河麒麟深度学习框架 KylinDLA\u003C\u002Fh4>\u003Cul>\n\u003Cli>统一适配多版本 PyTorch，兼容英伟达、AMD、国产各类异构加速卡，抹平框架版本差异。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw18.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=5ji9VKzvKL0417kKezFrBios5ZY%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"2-面向国产-cpu-gpu-的算子优化\">\u003Ca class=\"anchor\" href=\"#2-面向国产-cpu-gpu-的算子优化\" aria-label=\"跳转到标题：2.面向国产 CPU+GPU 的算子优化\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>2.面向国产 CPU+GPU 的算子优化\u003C\u002Fh4>\u003Cul>\n\u003Cli>算子融合：合并连续计算，削减 HBM 访存读写；\u003C\u002Fli>\n\u003Cli>内存调度：KV Cache 分页、张量复用，降低峰值显存占用；\u003C\u002Fli>\n\u003Cli>CPU\u002FGPU 协同分工，均衡负载；\u003C\u002Fli>\n\u003Cli>已完成多个国产 GPU 专用算子深度调优。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw19.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=V5sXT6NZzEgr3ezozb7y8G8SA3M%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"3-异构加速卡统一通信库\">\u003Ca class=\"anchor\" href=\"#3-异构加速卡统一通信库\" aria-label=\"跳转到标题：3. 异构加速卡统一通信库\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>3. 异构加速卡统一通信库\u003C\u002Fh4>\u003Cp>分层模块化设计，分为策略层、模型层、机制层：\u003C\u002Fp>\n\u003Cul>\n\u003Cli>同芯片：调用厂商原生通信接口；\u003C\u002Fli>\n\u003Cli>跨异构芯片： 通过 RDMA 等高速互联技术实现不同计算设备间的数据传输与通信优化；\u003C\u002Fli>\n\u003Cli>内置 AllReduce \u002F AllGather 集合通信优化，支持多卡混合训练、推理。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw20.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=NtyWQzHNMzHfu%2BMHIb3tEKoOYCw%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"4-麒麟智算操作系统\">\u003Ca class=\"anchor\" href=\"#4-麒麟智算操作系统\" aria-label=\"跳转到标题：4. 麒麟智算操作系统\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>4. 麒麟智算操作系统\u003C\u002Fh4>\u003Cul>\n\u003Cli>独立开发沐曦、天枢、海光等芯片专属算子库，支持 FP 8 \u002F INT 8 量化、GQA \u002F MQA 注意力、矩阵乘，7B \u002F 72 B 大模型推理性能较原厂提升显著。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw21.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Cx9vQeq6yOPiV5FKNrG0t5%2FNSyY%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"上层模型生态适配\">\u003Ca class=\"anchor\" href=\"#上层模型生态适配\" aria-label=\"跳转到标题：上层模型生态适配\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>上层模型生态适配\u003C\u002Fh3>\u003Cul>\n\u003Cli>已适配大语言、多模态视觉、语音、推荐等200+ 主流模型，配套智算生态网站一键部署。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw22.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=labxNqg5elbwUHbqT1FOlR%2BZH58%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"基于全国产智算底座的一体机解决方案\">\u003Ca class=\"anchor\" href=\"#基于全国产智算底座的一体机解决方案\" aria-label=\"跳转到标题：基于全国产智算底座的一体机解决方案\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>基于全国产智算底座的一体机解决方案\u003C\u002Fh3>\u003Col>\n\u003Cli>\u003Cstrong>数字员工一体机\u003C\u002Fstrong>：问答、文档处理、翻译、会议多场景封装，私有化部署；\u003C\u002Fli>\n\u003Cli>\u003Cstrong>AI 教学一体机\u003C\u002Fstrong>：配套训练推理实操题目，适配训练营课程开发调试；\u003C\u002Fli>\n\u003Cli>\u003Cstrong>编程助手一体机\u003C\u002Fstrong>：内网代码生成、工作流自动化、代码评审；\u003C\u002Fli>\n\u003Cli>\u003Cstrong>智慧客服一体机\u003C\u002Fstrong>：专属知识库、多轮对话、工单自动生成。\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw23.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=HiFjaCxy31XGRO13iXlXgYFCS7E%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch2 id=\"端侧-ai-软件栈和-agent-运行时系统\">\u003Ca class=\"anchor\" href=\"#端侧-ai-软件栈和-agent-运行时系统\" aria-label=\"跳转到标题：端侧 AI 软件栈和 Agent 运行时系统\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>端侧 AI 软件栈和 Agent 运行时系统\u003C\u002Fh2>\u003Ch3 id=\"端侧-ai-架构\">\u003Ca class=\"anchor\" href=\"#端侧-ai-架构\" aria-label=\"跳转到标题：端侧 AI 架构\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>端侧 AI 架构\u003C\u002Fh3>\u003Cp>端侧硬件约束：算力有限、带宽低、显存小，侧重功耗与算力平衡。\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw24.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=I1vcKUFgYAVzWa0e%2F6%2FhnTXJW2A%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"端侧-moe-模型推理优化\">\u003Ca class=\"anchor\" href=\"#端侧-moe-模型推理优化\" aria-label=\"跳转到标题：端侧 MoE 模型推理优化\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>端侧 MoE 模型推理优化\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw25.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=9shS08qzFuiIv1WazfQ7%2BEccboA%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"1-降低路由-topk-减少专家激活数量-降低计算量\">\u003Ca class=\"anchor\" href=\"#1-降低路由-topk-减少专家激活数量-降低计算量\" aria-label=\"跳转到标题：1. 降低路由 TopK：减少专家激活数量，降低计算量；\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>1. 降低路由 TopK：减少专家激活数量，降低计算量；\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw26.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=71NJqJh7wEFtp8t5BA815cMlBks%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"2-缩小单个专家\">\u003Ca class=\"anchor\" href=\"#2-缩小单个专家\" aria-label=\"跳转到标题：2. 缩小单个专家：\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>2. 缩小单个专家：\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw27.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=avI0ATH%2BMXjMvop5c35oHjidARU%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"3-路由算法优化-简单输入少激活专家-复杂输入多专家协同\">\u003Ca class=\"anchor\" href=\"#3-路由算法优化-简单输入少激活专家-复杂输入多专家协同\" aria-label=\"跳转到标题：3. 路由算法优化：简单输入少激活专家，复杂输入多专家协同；\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>3. 路由算法优化：简单输入少激活专家，复杂输入多专家协同；\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw28.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=uPUa1WrAXNCNJGVcpy9mvUjfoGg%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"4-推理部署优化-专家卸载-推测解码\">\u003Ca class=\"anchor\" href=\"#4-推理部署优化-专家卸载-推测解码\" aria-label=\"跳转到标题：4. 推理部署优化：专家卸载；推测解码。\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>4. 推理部署优化：专家卸载；推测解码。\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw29.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=vmUIceDien8GBQib9Hg1ha%2BjlRI%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"推理性能-profiler-分析工具\">\u003Ca class=\"anchor\" href=\"#推理性能-profiler-分析工具\" aria-label=\"跳转到标题：推理性能 Profiler 分析工具\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>推理性能 Profiler 分析工具\u003C\u002Fh3>\u003Cp>算子级追踪 Kernel 耗时、访存开销，自动给出剪枝、量化、算子融合优化建议，适配全系列国产 NPU \u002F GPU。\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw30.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=CiC43wVxaZ%2FNVOes3Dnp3c3R7S8%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"注意力算子优化-面向-gqa-mqa\">\u003Ca class=\"anchor\" href=\"#注意力算子优化-面向-gqa-mqa\" aria-label=\"跳转到标题：注意力算子优化（面向 GQA \u002F MQA）\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>注意力算子优化（面向 GQA \u002F MQA）\u003C\u002Fh3>\u003Cp>重构内存排布，合并单 Block 内多头计算，提升 Warp 归约效率，访存下降，整体性能提升 20% +。\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw31.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=fc9ZA9NctSrKRaKIZN5wirLRz58%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"sampler-采样算子融合\">\u003Ca class=\"anchor\" href=\"#sampler-采样算子融合\" aria-label=\"跳转到标题：Sampler 采样算子融合\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>Sampler 采样算子融合\u003C\u002Fh3>\u003Cp>整合 TopK \u002F TopP \u002F Temperature 全套逻辑，消除多次 Load \u002F Store 访存冗余，Batch 越大加速效果越明显。\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw32.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=xgULZwHUE%2B5QhZyu4NyNpasgkF4%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"量化访存优化-gptq-awq\">\u003Ca class=\"anchor\" href=\"#量化访存优化-gptq-awq\" aria-label=\"跳转到标题：量化访存优化（GPTQ、AWQ）\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>量化访存优化（GPTQ、AWQ）\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw33.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=nMKDz4L5HWb9Vq%2FmYsGhCRbSAnI%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_100_zw34.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=HNeZz%2FdogSl%2BQXhf6ZGfz78Qd%2F0%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch2 id=\"加入训练营\">\u003Ca class=\"anchor\" href=\"#加入训练营\" aria-label=\"跳转到标题：加入训练营\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>加入训练营\u003C\u002Fh2>\u003Cp>📺\u003Cstrong>完整课程内容，请观看直播或查看课程回放：\u003C\u002Fstrong>\u003C\u002Fp>\n\u003Cp>\u003Cstrong>InfiniTensor 官网\u003C\u002Fstrong>： \u003Ca href=\"https:\u002F\u002Fwww.infinitensor.com\">https:\u002F\u002Fwww.infinitensor.com\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp>\u003Cstrong>B站直播\u003C\u002Fstrong>： InfiniTensor 官方直播间\u003C\u002Fp>\n\u003Cp>\u003Cstrong>视频号直播\u003C\u002Fstrong>： InfiniTensor 视频号预约\u003C\u002Fp>\n\u003Cp>\u003Cstrong>答疑交流\u003C\u002Fstrong>： 训练营官方社群⬇️\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002Fqun_1072.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=dUe%2BM0SLl6KCrp97rPdQhphT%2Bck%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch2 id=\"关注与交流\">\u003Ca class=\"anchor\" href=\"#关注与交流\" aria-label=\"跳转到标题：关注与交流\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>关注与交流\u003C\u002Fh2>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002Fportal_news\u002Fnews_ewm02.jpg?Expires=1792750032&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=6FVLIDG%2FuTYOS25cEBcGpul%2Bzno%3D\" alt=\"\">\u003C\u002Fp>\n",[128,130,132,134,136,138,140,143,145,149,152,155,158,160,162,165,168,171,174,177,180,183,186,189,192,195,197],{"id":129,"text":129,"level":110},"麒麟软件与国产操作系统产品体系",{"id":131,"text":131,"level":9},"麒麟发展历程",{"id":133,"text":133,"level":9},"核心操作系统产品",{"id":135,"text":135,"level":9},"生态适配规模",{"id":137,"text":137,"level":110},"智算操作系统和智算软件栈",{"id":139,"text":139,"level":9},"国家政策环境",{"id":141,"text":142,"level":9},"国产-ai-软件栈发展现状","国产 AI 软件栈发展现状",{"id":144,"text":144,"level":9},"智算操作系统核心目标",{"id":146,"text":147,"level":148},"1-银河麒麟深度学习框架-kylindla","1.银河麒麟深度学习框架 KylinDLA",4,{"id":150,"text":151,"level":148},"2-面向国产-cpu-gpu-的算子优化","2.面向国产 CPU+GPU 的算子优化",{"id":153,"text":154,"level":148},"3-异构加速卡统一通信库","3. 异构加速卡统一通信库",{"id":156,"text":157,"level":148},"4-麒麟智算操作系统","4. 麒麟智算操作系统",{"id":159,"text":159,"level":9},"上层模型生态适配",{"id":161,"text":161,"level":9},"基于全国产智算底座的一体机解决方案",{"id":163,"text":164,"level":110},"端侧-ai-软件栈和-agent-运行时系统","端侧 AI 软件栈和 Agent 运行时系统",{"id":166,"text":167,"level":9},"端侧-ai-架构","端侧 AI 架构",{"id":169,"text":170,"level":9},"端侧-moe-模型推理优化","端侧 MoE 模型推理优化",{"id":172,"text":173,"level":148},"1-降低路由-topk-减少专家激活数量-降低计算量","1. 降低路由 TopK：减少专家激活数量，降低计算量；",{"id":175,"text":176,"level":148},"2-缩小单个专家","2. 缩小单个专家：",{"id":178,"text":179,"level":148},"3-路由算法优化-简单输入少激活专家-复杂输入多专家协同","3. 路由算法优化：简单输入少激活专家，复杂输入多专家协同；",{"id":181,"text":182,"level":148},"4-推理部署优化-专家卸载-推测解码","4. 推理部署优化：专家卸载；推测解码。",{"id":184,"text":185,"level":9},"推理性能-profiler-分析工具","推理性能 Profiler 分析工具",{"id":187,"text":188,"level":9},"注意力算子优化-面向-gqa-mqa","注意力算子优化（面向 GQA \u002F MQA）",{"id":190,"text":191,"level":9},"sampler-采样算子融合","Sampler 采样算子融合",{"id":193,"text":194,"level":9},"量化访存优化-gptq-awq","量化访存优化（GPTQ、AWQ）",{"id":196,"text":196,"level":110},"加入训练营",{"id":198,"text":198,"level":110},"关注与交流"]