[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"site-home-live-entry-setting":3,"site-header-contests":8,"site-header-camps":49,"study-material-detail-study-51":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=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=YuOgm%2FgmD4nuySjwNMl5nm0fmKI%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=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Kxv2UTqtU0%2FwvYQCDERqdg0W6E4%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=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=nIij3MTYWjwdhBM00gMTqmyLnUM%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_51_fm.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=r%2FCVMARKF4XtsfSL1nxwKsewWiE%3D","InfiniTensor,冬季训练营,强化学习,DeepSeek,训练系统","围绕强化学习与 DeepSeek 训练思路展开，梳理关键原理、训练策略与实战要点。","2026-10-22 20:38:11",2,"InfiniTensor 冬季训练营训练系统第五课《强化学习及 DeepSeek》","![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw01.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=91dMnADmTcebL3D9vIm9VkZl%2BCU%3D)\r\n\r\n\r\n\r\n2026 年 2 月 2 日，2025 冬季 InfiniTensor 大模型与人工智能系统训练营训练系统系列第五课《强化学习及 DeepSeek》开讲。\r\n\r\n本节课第一部分讲强化学习基础概念，第二部分介绍 DeepSeek 的一些主要架构以及涉及到的训练方法。\r\n\r\n## 强化学习基础概念\r\n### 监督学习与强化学习的对比\r\n#### 1. 监督学习：\r\n- 如同在老师指导下学习，提供大量带标签数据。\r\n- 特点：数据现成、样本独立、任务预测性。\r\n\r\n#### 2. 强化学习：\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw02.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=7y%2BppGFuQBtyZiWYYre8%2BD0B2Hg%3D)\r\n\r\n### 强化学习\r\n#### 1.要素：环境、智能体、状态、动作、奖励\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw03.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=XP74RBswJE7FhBkwsI2zBQuBT78%3D)\r\n\r\n#### 2.马尔可夫决策过程（MDP）：\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw04.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=N%2BmOH7I7YBocxY9BNN5cjhdBn%2Bo%3D)\r\n\r\n\r\n\r\n#### 3. 马尔可夫奖励过程（MRP）\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw05.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=SKICU8HQKloaQ79%2BspFHDPLPKNE%3D)\r\n\r\n- 要素：四元组(S, P, R, γ)，R 为奖励函数，γ 为折扣因子。\r\n\r\n#### 4. 价值函数：\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw06.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=k%2BIGEwer8D7F5zGrEYjrhrj7tNs%3D)\r\n\r\n#### 4. 贝尔曼方程：\r\n- 定义：价值函数的递归结构。\r\n- 应用：强化学习算法的理论基础。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw07.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Hmlzzu13ObxpmrKeXNzITzPz3N4%3D)\r\n\r\n\r\n#### 5. 马尔可夫决策过程（MDP）\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw08.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=i1LLSu7yc3fd58xQQDD3sGz2n50%3D)\r\n\r\n（1）定义：引进动作的马尔可夫奖励过程。\r\n（2）要素：五元组(S, A, P, R, γ)，A 为动作集合。\r\n（3）策略与价值函数：\r\n- 策略：智能体根据当前状态选择动作的函数。\r\n- 状态价值函数与动作价值函数：评估策略好坏的指标。\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw09.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=69kgmwaBAdyjvF%2FZW1nE%2Futx8EU%3D)\r\n\r\n（4）最优策略与贝尔曼最优方程：\r\n- 最优策略：最大化长期累积奖励的策略。\r\n- 贝尔曼最优方程：描述最优价值函数的递归结构。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw10.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=sqiMQNMxDmqb8%2F%2FSip0w869bB6c%3D)\r\n\r\n\r\n#### 6. Q-Learning 算法\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw11.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=FktMIIJLZuhyNlG0HRWAdxjmISo%3D)\r\n\r\n- 定义：帮助智能体学会最优策略的经典强化学习算法。\r\n- 核心思想：学会 Q 表，存储状态下做动作的长期奖励。\r\n- 更新公式：使用目标值与 TD 误差更新 Q 值。\r\n- 学习过程：初始化 Q 表、观察状态、选择动作、更新 Q 值。\r\n\r\n#### 7. DQN 算法\r\n- 定义：在 Q-Learning 基础上引入神经网络的算法。\r\n- 特点：使用神经网络近似 Q 函数，适用于连续或大规模状态空间。\r\n- 关键模块：经验回放与目标网络。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw12.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=y2AW0E00hWJrNa%2Bqu5lZr1%2F6smQ%3D)\r\n\r\n\r\n#### 8. 基于策略的方法与策略梯度\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw13.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=A6W7gH6lSXNShZ31oNt1tYMrckc%3D)\r\n\r\n- 定义：直接学习目标策略的方法。\r\n- 策略梯度：参数化策略，通过优化参数最大化期望回报。\r\n\r\n#### 9. Actor-Critic 算法\r\n- 定义：结合基于价值与基于策略的方法。\r\n- 核心：Actor 负责选择动作，Critic 负责评估策略表现。\r\n- 关键概念：优势函数、TD 误差、策略更新方式。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw14.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=W5t7uoM%2BEEDDZ78xsdHSdm0sksI%3D)\r\n\r\n#### 10. PPO 算法\r\n- 定义：限制梯度更新幅度的策略优化算法。\r\n- 核心思想：裁剪机制与\"小步快跑\"策略更新方式。\r\n- 应用：大语言模型微调，通过生成响应、获取反馈、调整模型流程优化性能。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw15.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=WKmG6FeKT9273OkyZdvl7lfd9L0%3D)\r\n\r\n\r\n\r\n## DeepSeek 架构及训练方法\r\n### DeepSeek 架构演进概述\r\n- 时间线：从 2024 年底到 2025 年底的架构版本演进。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw16.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=y9%2FQIeZiPGaEyqVePMHsz0P7pm8%3D)\r\n\r\n\r\n### DeepSeek V3 架构\r\n- 特点：MoE 架构，超大规模参数但稀疏激活。\r\n- 核心设计：MLA 与 MoE，降低 Attention 计算显存占用。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw17.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=F%2B8mPvo8JI2ilYV3W25VEAxO9Bg%3D)\r\n\r\n### DeepSeek R1 架构\r\n- 基础：基于 DeepSeek V3。\r\n- 核心差异：训练目标转向推理能力，引入 RLVR 训练方法。\r\n- RLVR：基于可验证反馈信号的强化学习方法。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw18.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=JR1tuSAu5bfJUXUMVsBUXDIvEag%3D)\r\n\r\n### DeepSeek V3.2 架构\r\n- 基础：保留 MLA 与 MOE 基本架构。\r\n- 新增模块：DSA（DeepSeek Sparse Attention），降低长上下文推理计算与显存成本。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw19.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=BYuJb8HXl1z7xxEPc1IYfqN20JA%3D)\r\n\r\n- DSA 计算流程：\r\nToken embeddings 输入\r\nLighting Indexer 模块计算相关性\r\nTop-K Selector 保留得分最高历史 token\r\n构建 Sparse Attention Mask\r\n在稀疏 mark 上执行注意力计算。\r\n\r\n\r\n![2025冬季训练营课程表](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_kcb.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=WE6ZvjevfVWwBzWNiOLG517R7Kc%3D)\r\n\r\n### 💡 学习贴士\r\n更多细节及完整课程内容，请观看直播或查看课程回放。\r\n\r\n链接：https:\u002F\u002Fwww.infinitensor.com\u002Fcamp\u002Fwinter2025\r\n\r\n### 📚 训练营报名持续进行中\r\n扫码报名与加入群聊，或点击 **「阅读原文」** 直达官网报名\r\n\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_ewm.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=7X9%2Bd0L8rZJpZVU8z9o6IgBjCBo%3D)",{"publishTime":7,"createTime":109,"canonicalPath":114,"id":115,"title":116},"\u002Fstudy-material\u002Fstudy-50","study-50","InfiniTensor 冬季训练营 CUDA 编程第七课《量化与工业级调优部署》",58,"\u002Fstudy-material\u002Fstudy-51","study-51","markdown",{"publishTime":7,"createTime":109,"canonicalPath":122,"id":123,"title":124},"\u002Fstudy-material\u002Fstudy-52","study-52","InfiniTensor 冬季训练营推理引擎第七课《InfiniLM》",{"htmlContent":126,"headings":127},"\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw01.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=91dMnADmTcebL3D9vIm9VkZl%2BCU%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>2026 年 2 月 2 日，2025 冬季 InfiniTensor 大模型与人工智能系统训练营训练系统系列第五课《强化学习及 DeepSeek》开讲。\u003C\u002Fp>\n\u003Cp>本节课第一部分讲强化学习基础概念，第二部分介绍 DeepSeek 的一些主要架构以及涉及到的训练方法。\u003C\u002Fp>\n\u003Ch2 id=\"强化学习基础概念\">\u003Ca class=\"anchor\" href=\"#强化学习基础概念\" aria-label=\"跳转到标题：强化学习基础概念\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>强化学习基础概念\u003C\u002Fh2>\u003Ch3 id=\"监督学习与强化学习的对比\">\u003Ca class=\"anchor\" href=\"#监督学习与强化学习的对比\" aria-label=\"跳转到标题：监督学习与强化学习的对比\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>监督学习与强化学习的对比\u003C\u002Fh3>\u003Ch4 id=\"1-监督学习\">\u003Ca class=\"anchor\" href=\"#1-监督学习\" aria-label=\"跳转到标题：1. 监督学习：\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>1. 监督学习：\u003C\u002Fh4>\u003Cul>\n\u003Cli>如同在老师指导下学习，提供大量带标签数据。\u003C\u002Fli>\n\u003Cli>特点：数据现成、样本独立、任务预测性。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch4 id=\"2-强化学习\">\u003Ca class=\"anchor\" href=\"#2-强化学习\" aria-label=\"跳转到标题：2. 强化学习：\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>2. 强化学习：\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw02.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=7y%2BppGFuQBtyZiWYYre8%2BD0B2Hg%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"强化学习\">\u003Ca class=\"anchor\" href=\"#强化学习\" aria-label=\"跳转到标题：强化学习\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>强化学习\u003C\u002Fh3>\u003Ch4 id=\"1-要素-环境-智能体-状态-动作-奖励\">\u003Ca class=\"anchor\" href=\"#1-要素-环境-智能体-状态-动作-奖励\" aria-label=\"跳转到标题：1.要素：环境、智能体、状态、动作、奖励\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>1.要素：环境、智能体、状态、动作、奖励\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw03.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=XP74RBswJE7FhBkwsI2zBQuBT78%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"2-马尔可夫决策过程-mdp\">\u003Ca class=\"anchor\" href=\"#2-马尔可夫决策过程-mdp\" aria-label=\"跳转到标题：2.马尔可夫决策过程（MDP）：\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>2.马尔可夫决策过程（MDP）：\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw04.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=N%2BmOH7I7YBocxY9BNN5cjhdBn%2Bo%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"3-马尔可夫奖励过程-mrp\">\u003Ca class=\"anchor\" href=\"#3-马尔可夫奖励过程-mrp\" aria-label=\"跳转到标题：3. 马尔可夫奖励过程（MRP）\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>3. 马尔可夫奖励过程（MRP）\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw05.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=SKICU8HQKloaQ79%2BspFHDPLPKNE%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cul>\n\u003Cli>要素：四元组(S, P, R, γ)，R 为奖励函数，γ 为折扣因子。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch4 id=\"4-价值函数\">\u003Ca class=\"anchor\" href=\"#4-价值函数\" aria-label=\"跳转到标题：4. 价值函数：\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>4. 价值函数：\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw06.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=k%2BIGEwer8D7F5zGrEYjrhrj7tNs%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"4-贝尔曼方程\">\u003Ca class=\"anchor\" href=\"#4-贝尔曼方程\" aria-label=\"跳转到标题：4. 贝尔曼方程：\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>4. 贝尔曼方程：\u003C\u002Fh4>\u003Cul>\n\u003Cli>定义：价值函数的递归结构。\u003C\u002Fli>\n\u003Cli>应用：强化学习算法的理论基础。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw07.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Hmlzzu13ObxpmrKeXNzITzPz3N4%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"5-马尔可夫决策过程-mdp\">\u003Ca class=\"anchor\" href=\"#5-马尔可夫决策过程-mdp\" aria-label=\"跳转到标题：5. 马尔可夫决策过程（MDP）\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>5. 马尔可夫决策过程（MDP）\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw08.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=i1LLSu7yc3fd58xQQDD3sGz2n50%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>（1）定义：引进动作的马尔可夫奖励过程。\u003Cbr>（2）要素：五元组(S, A, P, R, γ)，A 为动作集合。\u003Cbr>（3）策略与价值函数：\u003C\u002Fp>\n\u003Cul>\n\u003Cli>策略：智能体根据当前状态选择动作的函数。\u003C\u002Fli>\n\u003Cli>状态价值函数与动作价值函数：评估策略好坏的指标。\u003Cbr>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw09.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=69kgmwaBAdyjvF%2FZW1nE%2Futx8EU%3D\" alt=\"\">\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>（4）最优策略与贝尔曼最优方程：\u003C\u002Fp>\n\u003Cul>\n\u003Cli>最优策略：最大化长期累积奖励的策略。\u003C\u002Fli>\n\u003Cli>贝尔曼最优方程：描述最优价值函数的递归结构。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw10.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=sqiMQNMxDmqb8%2F%2FSip0w869bB6c%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"6-q-learning-算法\">\u003Ca class=\"anchor\" href=\"#6-q-learning-算法\" aria-label=\"跳转到标题：6. Q-Learning 算法\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>6. Q-Learning 算法\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw11.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=FktMIIJLZuhyNlG0HRWAdxjmISo%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cul>\n\u003Cli>定义：帮助智能体学会最优策略的经典强化学习算法。\u003C\u002Fli>\n\u003Cli>核心思想：学会 Q 表，存储状态下做动作的长期奖励。\u003C\u002Fli>\n\u003Cli>更新公式：使用目标值与 TD 误差更新 Q 值。\u003C\u002Fli>\n\u003Cli>学习过程：初始化 Q 表、观察状态、选择动作、更新 Q 值。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch4 id=\"7-dqn-算法\">\u003Ca class=\"anchor\" href=\"#7-dqn-算法\" aria-label=\"跳转到标题：7. DQN 算法\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>7. DQN 算法\u003C\u002Fh4>\u003Cul>\n\u003Cli>定义：在 Q-Learning 基础上引入神经网络的算法。\u003C\u002Fli>\n\u003Cli>特点：使用神经网络近似 Q 函数，适用于连续或大规模状态空间。\u003C\u002Fli>\n\u003Cli>关键模块：经验回放与目标网络。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw12.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=y2AW0E00hWJrNa%2Bqu5lZr1%2F6smQ%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"8-基于策略的方法与策略梯度\">\u003Ca class=\"anchor\" href=\"#8-基于策略的方法与策略梯度\" aria-label=\"跳转到标题：8. 基于策略的方法与策略梯度\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>8. 基于策略的方法与策略梯度\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw13.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=A6W7gH6lSXNShZ31oNt1tYMrckc%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cul>\n\u003Cli>定义：直接学习目标策略的方法。\u003C\u002Fli>\n\u003Cli>策略梯度：参数化策略，通过优化参数最大化期望回报。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch4 id=\"9-actor-critic-算法\">\u003Ca class=\"anchor\" href=\"#9-actor-critic-算法\" aria-label=\"跳转到标题：9. Actor-Critic 算法\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>9. Actor-Critic 算法\u003C\u002Fh4>\u003Cul>\n\u003Cli>定义：结合基于价值与基于策略的方法。\u003C\u002Fli>\n\u003Cli>核心：Actor 负责选择动作，Critic 负责评估策略表现。\u003C\u002Fli>\n\u003Cli>关键概念：优势函数、TD 误差、策略更新方式。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw14.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=W5t7uoM%2BEEDDZ78xsdHSdm0sksI%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"10-ppo-算法\">\u003Ca class=\"anchor\" href=\"#10-ppo-算法\" aria-label=\"跳转到标题：10. PPO 算法\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>10. PPO 算法\u003C\u002Fh4>\u003Cul>\n\u003Cli>定义：限制梯度更新幅度的策略优化算法。\u003C\u002Fli>\n\u003Cli>核心思想：裁剪机制与"小步快跑"策略更新方式。\u003C\u002Fli>\n\u003Cli>应用：大语言模型微调，通过生成响应、获取反馈、调整模型流程优化性能。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw15.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=WKmG6FeKT9273OkyZdvl7lfd9L0%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch2 id=\"deepseek-架构及训练方法\">\u003Ca class=\"anchor\" href=\"#deepseek-架构及训练方法\" aria-label=\"跳转到标题：DeepSeek 架构及训练方法\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>DeepSeek 架构及训练方法\u003C\u002Fh2>\u003Ch3 id=\"deepseek-架构演进概述\">\u003Ca class=\"anchor\" href=\"#deepseek-架构演进概述\" aria-label=\"跳转到标题：DeepSeek 架构演进概述\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>DeepSeek 架构演进概述\u003C\u002Fh3>\u003Cul>\n\u003Cli>时间线：从 2024 年底到 2025 年底的架构版本演进。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw16.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=y9%2FQIeZiPGaEyqVePMHsz0P7pm8%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"deepseek-v3-架构\">\u003Ca class=\"anchor\" href=\"#deepseek-v3-架构\" aria-label=\"跳转到标题：DeepSeek V3 架构\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>DeepSeek V3 架构\u003C\u002Fh3>\u003Cul>\n\u003Cli>特点：MoE 架构，超大规模参数但稀疏激活。\u003C\u002Fli>\n\u003Cli>核心设计：MLA 与 MoE，降低 Attention 计算显存占用。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw17.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=F%2B8mPvo8JI2ilYV3W25VEAxO9Bg%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"deepseek-r1-架构\">\u003Ca class=\"anchor\" href=\"#deepseek-r1-架构\" aria-label=\"跳转到标题：DeepSeek R1 架构\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>DeepSeek R1 架构\u003C\u002Fh3>\u003Cul>\n\u003Cli>基础：基于 DeepSeek V3。\u003C\u002Fli>\n\u003Cli>核心差异：训练目标转向推理能力，引入 RLVR 训练方法。\u003C\u002Fli>\n\u003Cli>RLVR：基于可验证反馈信号的强化学习方法。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw18.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=JR1tuSAu5bfJUXUMVsBUXDIvEag%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"deepseek-v3-2-架构\">\u003Ca class=\"anchor\" href=\"#deepseek-v3-2-架构\" aria-label=\"跳转到标题：DeepSeek V3.2 架构\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>DeepSeek V3.2 架构\u003C\u002Fh3>\u003Cul>\n\u003Cli>基础：保留 MLA 与 MOE 基本架构。\u003C\u002Fli>\n\u003Cli>新增模块：DSA（DeepSeek Sparse Attention），降低长上下文推理计算与显存成本。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_51_zw19.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=BYuJb8HXl1z7xxEPc1IYfqN20JA%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cul>\n\u003Cli>DSA 计算流程：\u003Cbr>Token embeddings 输入\u003Cbr>Lighting Indexer 模块计算相关性\u003Cbr>Top-K Selector 保留得分最高历史 token\u003Cbr>构建 Sparse Attention Mask\u003Cbr>在稀疏 mark 上执行注意力计算。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_kcb.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=WE6ZvjevfVWwBzWNiOLG517R7Kc%3D\" alt=\"2025冬季训练营课程表\">\u003C\u002Fp>\n\u003Ch3 id=\"学习贴士\">\u003Ca class=\"anchor\" href=\"#学习贴士\" aria-label=\"跳转到标题：💡 学习贴士\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>💡 学习贴士\u003C\u002Fh3>\u003Cp>更多细节及完整课程内容，请观看直播或查看课程回放。\u003C\u002Fp>\n\u003Cp>链接：\u003Ca href=\"https:\u002F\u002Fwww.infinitensor.com\u002Fcamp\u002Fwinter2025\">https:\u002F\u002Fwww.infinitensor.com\u002Fcamp\u002Fwinter2025\u003C\u002Fa>\u003C\u002Fp>\n\u003Ch3 id=\"训练营报名持续进行中\">\u003Ca class=\"anchor\" href=\"#训练营报名持续进行中\" aria-label=\"跳转到标题：📚 训练营报名持续进行中\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>📚 训练营报名持续进行中\u003C\u002Fh3>\u003Cp>扫码报名与加入群聊，或点击 \u003Cstrong>「阅读原文」\u003C\u002Fstrong> 直达官网报名\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_ewm.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=7X9%2Bd0L8rZJpZVU8z9o6IgBjCBo%3D\" alt=\"\">\u003C\u002Fp>\n",[128,130,132,136,139,141,144,147,150,153,156,159,162,165,168,171,174,177,180,183,186,189,192],{"id":129,"text":129,"level":110},"强化学习基础概念",{"id":131,"text":131,"level":9},"监督学习与强化学习的对比",{"id":133,"text":134,"level":135},"1-监督学习","1. 监督学习：",4,{"id":137,"text":138,"level":135},"2-强化学习","2. 强化学习：",{"id":140,"text":140,"level":9},"强化学习",{"id":142,"text":143,"level":135},"1-要素-环境-智能体-状态-动作-奖励","1.要素：环境、智能体、状态、动作、奖励",{"id":145,"text":146,"level":135},"2-马尔可夫决策过程-mdp","2.马尔可夫决策过程（MDP）：",{"id":148,"text":149,"level":135},"3-马尔可夫奖励过程-mrp","3. 马尔可夫奖励过程（MRP）",{"id":151,"text":152,"level":135},"4-价值函数","4. 价值函数：",{"id":154,"text":155,"level":135},"4-贝尔曼方程","4. 贝尔曼方程：",{"id":157,"text":158,"level":135},"5-马尔可夫决策过程-mdp","5. 马尔可夫决策过程（MDP）",{"id":160,"text":161,"level":135},"6-q-learning-算法","6. Q-Learning 算法",{"id":163,"text":164,"level":135},"7-dqn-算法","7. DQN 算法",{"id":166,"text":167,"level":135},"8-基于策略的方法与策略梯度","8. 基于策略的方法与策略梯度",{"id":169,"text":170,"level":135},"9-actor-critic-算法","9. Actor-Critic 算法",{"id":172,"text":173,"level":135},"10-ppo-算法","10. PPO 算法",{"id":175,"text":176,"level":110},"deepseek-架构及训练方法","DeepSeek 架构及训练方法",{"id":178,"text":179,"level":9},"deepseek-架构演进概述","DeepSeek 架构演进概述",{"id":181,"text":182,"level":9},"deepseek-v3-架构","DeepSeek V3 架构",{"id":184,"text":185,"level":9},"deepseek-r1-架构","DeepSeek R1 架构",{"id":187,"text":188,"level":9},"deepseek-v3-2-架构","DeepSeek V3.2 架构",{"id":190,"text":191,"level":9},"学习贴士","💡 学习贴士",{"id":193,"text":194,"level":9},"训练营报名持续进行中","📚 训练营报名持续进行中"]