[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"site-home-live-entry-setting":3,"site-header-contests":8,"site-header-camps":49,"study-material-detail-study-107":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=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=739batJ4%2B57Fju5PIvpwAi5wwJ4%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=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=TtQRCdsl2%2Blbx4NBmB3M41mPGDA%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=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=q4JuhVDAFMLSebXnpwWGpJRLRAw%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":124},{"img":106,"keywords":107,"treeType":44,"description":108,"showStatus":44,"updateTime":109,"type":110,"title":111,"content":112,"nextArticle":113,"createTime":109,"sortOrder":9,"canonicalPath":117,"id":118,"contentType":119,"prevArticle":120},"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_fm.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=bJKxdP5sdFbz6N0Ae%2BUwZ3J%2B9SE%3D","九齿 DSL,算子开发,Triton,TileLang,多范式算子,并行编程","多范式算子开发实战营《九齿的设计思想与核心编程范式》深入解析九齿 DSL 的设计思想与核心编程范式，横向对比 Triton、TileLang 的实现差异。详解 Triton 四大核心 API 及 Block 级编程模型，并通过向量加法实战演示九齿 Arrangement 与 Application 分离的设计精髓、Make API 组合逻辑及双重可视化工具，帮助开发者掌握张量级抽象的并行编程思维。","2026-10-22 20:38:11",2,"多范式算子开发实战营: 《九齿的设计思想与核心编程范式》课程回顾","9 月 15 日晚，多范式算子开发实战营邀请**九齿 DSL 技术负责人黄嘉成老师**，带来《九齿的设计思想与核心编程范式》。\r\n\r\n本节课将介绍九齿的设计思想与核心编程范式，并结合九齿、Triton、TileLang 等多种 DSL，介绍不同 DSL 的设计理念及适用场景，还将结合概念验证阶段的实际开发需求，帮助大家了解 DSL 如何用于算子开发与硬件适配，以及不同开发方式之间的特点和差异。\r\n\r\n\r\n## 一、 领域特定语言（DSL）\r\n### 什么是 DSL？\r\n1.Domain Specific Language 的定义与优势\r\n\r\n2.AI\u002F深度学习领域的常见 DSL：Triton, TileLang, 九齿\r\n\r\n### 三者横向对比：以“向量加法”为例\r\n**1. 向量加法**：并行编程界的 \"Hello World\"\r\n\r\n**2. 代码风格差异**：\r\n- Triton：显式 Program ID + 指针\u002F索引运算\r\n- TileLang：Pythonic `with` 语法 + Tiling 控制\r\n- 九齿：Arrangement（排布）+ Application（应用）分离\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw01.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=doeTdVFcQHkrVbaN8BYFsjEN%2Bvs%3D)\r\n\r\n### 设计出发点\r\n**1. Triton**：屏蔽 CUDA 底层细节，保留 Block 级并行语义，跨平台性较好但仍有硬件差异\r\n\r\n**2. TileLang**：显式暴露存储层级与流水线控制，追求更底层的手动优化空间\r\n\r\n**3. 九齿**：进一步抽象掉指针与偏移量计算，追求更好的跨平台性与初学者友好度；以编译器优化换取高层抽象的性能损失\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw02.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=xfquEk8gWjQ%2BjSbj%2BfZstpLzFG0%3D)\r\n\r\n\r\n## 二、 Triton 编程基础\r\n### 为什么需要 Triton？\r\n- CUDA 的痛点：语法复杂、硬件门槛高、内存\u002F线程管理繁琐\r\n- Triton 的优势：Block-level 抽象、代码量少、快速达到专家级 80%-90% 性能\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw03.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=vbqvvoI%2F%2BrKUjmMxEeHc%2FDP3218%3D)\r\n\r\n\r\n### 核心概念：从 Thread 到 Block\r\n- 无需显式管理 Thread 级执行细节、Shared Memory 等底层资源\r\n- 聚焦 SM 间调度与 Block 级数据访问\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw04.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=fqmOkGDfokHim3r9J2Q21QkuDco%3D)\r\n\r\n### 四大核心 API\r\n- `program_id`：定位当前 Worker 身份\r\n- `arange`：生成递增向量，构造数据访问区域\r\n- `load`：带 Mask 边界检查的数据读取（Pointer + Mask + Other）\r\n- `store`：带 Mask 的安全数据写回\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw05.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=jcsgcg8TyD8lkXiNOXfqX%2BipekY%3D)\r\n\r\n\r\n\r\n## 三、 九齿（NineToothed）核心设计思想\r\n### 九齿是什么？\r\n- 提供**张量级抽象**的 DSL\r\n- 无指针、无 Program ID、无 Load\u002FStore，一切皆张量\r\n- 以 Triton 为主要编译后端（支持 TensorFlow\u002FCUDA 扩展中）\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw06.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=wmAVexrxs1MYplTweA%2FelzdWk0U%3D)\r\n\r\n### 为什么需要九齿？\r\n- PyTorch 等框架层：接口简洁但定制化难\r\n- Triton 等指针级 DSL：仍需并行思维转换，入门有门槛\r\n- **九齿定位**：让未接触并行编程的用户也能高效编写 Kernel\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw07.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=QVREhaahNKKWt74gHfUkjq7l8Qc%3D)\r\n\r\n\r\n### 核心范式\r\n- **Arrangement（排布）**：描述数据如何切分与分布（Tile, Expand, Squeeze 等原语）\r\n- **Application（应用）**：描述单个 Tile 上的具体计算逻辑\r\n\r\n\r\n### 代码对比：矩阵乘法（GEMM）\r\n- Triton：厚重的 PID 变换、Offset 计算、指针 Increment\r\n- 九齿：零指针运算，代码贴近算法自然语言描述\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw08.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=8YqXGfXucY5VIM0T5afbaXfOCaU%3D)\r\n\r\n\r\n## 四、 DSL 的实际用途\r\n### Kernel vs Operator 的概念区分\r\n- Kernel：纯计算单元，不负责内存分配与通信\r\n- Operator：包含 Kernel 调用、内存管理、通信等广义算子\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw09.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=uCKG%2Fwtbt17sIh3FuXwCXt5ptCU%3D)\r\n\r\n### 核心应用场景\r\n- **现有算子融合\u002F优化**：如 FlashAttention，减少内存读写开销\r\n- **新算子敏捷概念验证（PoC）**：快速搭建，验证有效性后再投入底层优化\r\n- 新模型伴随新算子，DSL 加速“模型-算子”协同迭代\r\n\r\n## 五、 向量加法\r\n### Triton 实现\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw10.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=g4oiVmcr%2FnoG%2FxJi9l41VNg5Ijs%3D)\r\n\r\n\r\n### 九齿实现\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw11.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=yhl72bgcCaoqDQFNO1uoc2ygjlw%3D)\r\n\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw12.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=4umVuqOoHbCHL6VRdNmzXxd6ANk%3D)\r\n\r\n\r\n### 范式推广：从向量加法到 GEMM\u002FFlashAttention\r\n- 万能公式：若干 Tensor → Arrangement 生成多组 Tiles → Application 逐组 Apply\r\n- 核心理念：记住这三行关系，即掌握了九齿的所有复杂算子编写逻辑\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw13.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=CehC%2FBTrUEB8JgP2IB134KkZHzk%3D)\r\n\r\n### 双重可视化辅助理解\r\n- **嵌套张量可视化**：展示 Tile 后的多层结构（如长度 16 向量 → 外层 8×内层 2 的嵌套向量）\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw14.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=E42IWZUqouXl4R6hgJ9RV4MaBlQ%3D)\r\n\r\n\r\n- **排布可视化**：交互展示每组 Tile 与原张量的位置对应关系\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw15.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Qf4IDMP4OB%2FSVdybmQH2AYn%2B79M%3D)\r\n\r\n\r\n\r\n\r\n*完整代码示例与作业请访问课程仓库。如有疑问欢迎在 Issue 区交流讨论。*\r\n\r\n## 关注与交流\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002Fportal_news\u002Fnews_ewm02.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=4vIkDijirVOEl01W4C%2B3fyE9q2Y%3D)",{"publishTime":7,"createTime":109,"canonicalPath":114,"id":115,"title":116},"\u002Fstudy-material\u002Fstudy-106","study-106","「多范式算子开发实战营」开营首讲：《九源软件栈核心技术与开源生态》 课程回顾","\u002Fstudy-material\u002Fstudy-107","study-107","markdown",{"publishTime":7,"createTime":109,"canonicalPath":121,"id":122,"title":123},"\u002Fstudy-material\u002Fstudy-108","study-108","课程回顾 | 多范式算子开发实战营：《基于九齿复杂算子实现案例》",{"htmlContent":125,"headings":126},"\u003Cp>9 月 15 日晚，多范式算子开发实战营邀请\u003Cstrong>九齿 DSL 技术负责人黄嘉成老师\u003C\u002Fstrong>，带来《九齿的设计思想与核心编程范式》。\u003C\u002Fp>\n\u003Cp>本节课将介绍九齿的设计思想与核心编程范式，并结合九齿、Triton、TileLang 等多种 DSL，介绍不同 DSL 的设计理念及适用场景，还将结合概念验证阶段的实际开发需求，帮助大家了解 DSL 如何用于算子开发与硬件适配，以及不同开发方式之间的特点和差异。\u003C\u002Fp>\n\u003Ch2 id=\"一-领域特定语言-dsl\">\u003Ca class=\"anchor\" href=\"#一-领域特定语言-dsl\" aria-label=\"跳转到标题：一、 领域特定语言（DSL）\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>一、 领域特定语言（DSL）\u003C\u002Fh2>\u003Ch3 id=\"什么是-dsl\">\u003Ca class=\"anchor\" href=\"#什么是-dsl\" aria-label=\"跳转到标题：什么是 DSL？\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>什么是 DSL？\u003C\u002Fh3>\u003Cp>1.Domain Specific Language 的定义与优势\u003C\u002Fp>\n\u003Cp>2.AI\u002F深度学习领域的常见 DSL：Triton, TileLang, 九齿\u003C\u002Fp>\n\u003Ch3 id=\"三者横向对比-以-向量加法-为例\">\u003Ca class=\"anchor\" href=\"#三者横向对比-以-向量加法-为例\" aria-label=\"跳转到标题：三者横向对比：以“向量加法”为例\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>三者横向对比：以“向量加法”为例\u003C\u002Fh3>\u003Cp>\u003Cstrong>1. 向量加法\u003C\u002Fstrong>：并行编程界的 "Hello World"\u003C\u002Fp>\n\u003Cp>\u003Cstrong>2. 代码风格差异\u003C\u002Fstrong>：\u003C\u002Fp>\n\u003Cul>\n\u003Cli>Triton：显式 Program ID + 指针\u002F索引运算\u003C\u002Fli>\n\u003Cli>TileLang：Pythonic \u003Ccode>with\u003C\u002Fcode> 语法 + Tiling 控制\u003C\u002Fli>\n\u003Cli>九齿：Arrangement（排布）+ Application（应用）分离\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw01.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=doeTdVFcQHkrVbaN8BYFsjEN%2Bvs%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"设计出发点\">\u003Ca class=\"anchor\" href=\"#设计出发点\" aria-label=\"跳转到标题：设计出发点\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>设计出发点\u003C\u002Fh3>\u003Cp>\u003Cstrong>1. Triton\u003C\u002Fstrong>：屏蔽 CUDA 底层细节，保留 Block 级并行语义，跨平台性较好但仍有硬件差异\u003C\u002Fp>\n\u003Cp>\u003Cstrong>2. TileLang\u003C\u002Fstrong>：显式暴露存储层级与流水线控制，追求更底层的手动优化空间\u003C\u002Fp>\n\u003Cp>\u003Cstrong>3. 九齿\u003C\u002Fstrong>：进一步抽象掉指针与偏移量计算，追求更好的跨平台性与初学者友好度；以编译器优化换取高层抽象的性能损失\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw02.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=xfquEk8gWjQ%2BjSbj%2BfZstpLzFG0%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch2 id=\"二-triton-编程基础\">\u003Ca class=\"anchor\" href=\"#二-triton-编程基础\" aria-label=\"跳转到标题：二、 Triton 编程基础\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>二、 Triton 编程基础\u003C\u002Fh2>\u003Ch3 id=\"为什么需要-triton\">\u003Ca class=\"anchor\" href=\"#为什么需要-triton\" aria-label=\"跳转到标题：为什么需要 Triton？\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>为什么需要 Triton？\u003C\u002Fh3>\u003Cul>\n\u003Cli>CUDA 的痛点：语法复杂、硬件门槛高、内存\u002F线程管理繁琐\u003C\u002Fli>\n\u003Cli>Triton 的优势：Block-level 抽象、代码量少、快速达到专家级 80%-90% 性能\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw03.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=vbqvvoI%2F%2BrKUjmMxEeHc%2FDP3218%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"核心概念-从-thread-到-block\">\u003Ca class=\"anchor\" href=\"#核心概念-从-thread-到-block\" aria-label=\"跳转到标题：核心概念：从 Thread 到 Block\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>核心概念：从 Thread 到 Block\u003C\u002Fh3>\u003Cul>\n\u003Cli>无需显式管理 Thread 级执行细节、Shared Memory 等底层资源\u003C\u002Fli>\n\u003Cli>聚焦 SM 间调度与 Block 级数据访问\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw04.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=fqmOkGDfokHim3r9J2Q21QkuDco%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"四大核心-api\">\u003Ca class=\"anchor\" href=\"#四大核心-api\" aria-label=\"跳转到标题：四大核心 API\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>四大核心 API\u003C\u002Fh3>\u003Cul>\n\u003Cli>\u003Ccode>program_id\u003C\u002Fcode>：定位当前 Worker 身份\u003C\u002Fli>\n\u003Cli>\u003Ccode>arange\u003C\u002Fcode>：生成递增向量，构造数据访问区域\u003C\u002Fli>\n\u003Cli>\u003Ccode>load\u003C\u002Fcode>：带 Mask 边界检查的数据读取（Pointer + Mask + Other）\u003C\u002Fli>\n\u003Cli>\u003Ccode>store\u003C\u002Fcode>：带 Mask 的安全数据写回\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw05.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=jcsgcg8TyD8lkXiNOXfqX%2BipekY%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch2 id=\"三-九齿-ninetoothed-核心设计思想\">\u003Ca class=\"anchor\" href=\"#三-九齿-ninetoothed-核心设计思想\" aria-label=\"跳转到标题：三、 九齿（NineToothed）核心设计思想\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>三、 九齿（NineToothed）核心设计思想\u003C\u002Fh2>\u003Ch3 id=\"九齿是什么\">\u003Ca class=\"anchor\" href=\"#九齿是什么\" aria-label=\"跳转到标题：九齿是什么？\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>九齿是什么？\u003C\u002Fh3>\u003Cul>\n\u003Cli>提供\u003Cstrong>张量级抽象\u003C\u002Fstrong>的 DSL\u003C\u002Fli>\n\u003Cli>无指针、无 Program ID、无 Load\u002FStore，一切皆张量\u003C\u002Fli>\n\u003Cli>以 Triton 为主要编译后端（支持 TensorFlow\u002FCUDA 扩展中）\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw06.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=wmAVexrxs1MYplTweA%2FelzdWk0U%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"为什么需要九齿\">\u003Ca class=\"anchor\" href=\"#为什么需要九齿\" aria-label=\"跳转到标题：为什么需要九齿？\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>为什么需要九齿？\u003C\u002Fh3>\u003Cul>\n\u003Cli>PyTorch 等框架层：接口简洁但定制化难\u003C\u002Fli>\n\u003Cli>Triton 等指针级 DSL：仍需并行思维转换，入门有门槛\u003C\u002Fli>\n\u003Cli>\u003Cstrong>九齿定位\u003C\u002Fstrong>：让未接触并行编程的用户也能高效编写 Kernel\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw07.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=QVREhaahNKKWt74gHfUkjq7l8Qc%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"核心范式\">\u003Ca class=\"anchor\" href=\"#核心范式\" aria-label=\"跳转到标题：核心范式\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>核心范式\u003C\u002Fh3>\u003Cul>\n\u003Cli>\u003Cstrong>Arrangement（排布）\u003C\u002Fstrong>：描述数据如何切分与分布（Tile, Expand, Squeeze 等原语）\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Application（应用）\u003C\u002Fstrong>：描述单个 Tile 上的具体计算逻辑\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch3 id=\"代码对比-矩阵乘法-gemm\">\u003Ca class=\"anchor\" href=\"#代码对比-矩阵乘法-gemm\" aria-label=\"跳转到标题：代码对比：矩阵乘法（GEMM）\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>代码对比：矩阵乘法（GEMM）\u003C\u002Fh3>\u003Cul>\n\u003Cli>Triton：厚重的 PID 变换、Offset 计算、指针 Increment\u003C\u002Fli>\n\u003Cli>九齿：零指针运算，代码贴近算法自然语言描述\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw08.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=8YqXGfXucY5VIM0T5afbaXfOCaU%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch2 id=\"四-dsl-的实际用途\">\u003Ca class=\"anchor\" href=\"#四-dsl-的实际用途\" aria-label=\"跳转到标题：四、 DSL 的实际用途\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>四、 DSL 的实际用途\u003C\u002Fh2>\u003Ch3 id=\"kernel-vs-operator-的概念区分\">\u003Ca class=\"anchor\" href=\"#kernel-vs-operator-的概念区分\" aria-label=\"跳转到标题：Kernel vs Operator 的概念区分\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>Kernel vs Operator 的概念区分\u003C\u002Fh3>\u003Cul>\n\u003Cli>Kernel：纯计算单元，不负责内存分配与通信\u003C\u002Fli>\n\u003Cli>Operator：包含 Kernel 调用、内存管理、通信等广义算子\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw09.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=uCKG%2Fwtbt17sIh3FuXwCXt5ptCU%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"核心应用场景\">\u003Ca class=\"anchor\" href=\"#核心应用场景\" aria-label=\"跳转到标题：核心应用场景\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>核心应用场景\u003C\u002Fh3>\u003Cul>\n\u003Cli>\u003Cstrong>现有算子融合\u002F优化\u003C\u002Fstrong>：如 FlashAttention，减少内存读写开销\u003C\u002Fli>\n\u003Cli>\u003Cstrong>新算子敏捷概念验证（PoC）\u003C\u002Fstrong>：快速搭建，验证有效性后再投入底层优化\u003C\u002Fli>\n\u003Cli>新模型伴随新算子，DSL 加速“模型-算子”协同迭代\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2 id=\"五-向量加法\">\u003Ca class=\"anchor\" href=\"#五-向量加法\" aria-label=\"跳转到标题：五、 向量加法\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>五、 向量加法\u003C\u002Fh2>\u003Ch3 id=\"triton-实现\">\u003Ca class=\"anchor\" href=\"#triton-实现\" aria-label=\"跳转到标题：Triton 实现\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>Triton 实现\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw10.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=g4oiVmcr%2FnoG%2FxJi9l41VNg5Ijs%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"九齿实现\">\u003Ca class=\"anchor\" href=\"#九齿实现\" aria-label=\"跳转到标题：九齿实现\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>九齿实现\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw11.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=yhl72bgcCaoqDQFNO1uoc2ygjlw%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw12.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=4umVuqOoHbCHL6VRdNmzXxd6ANk%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"范式推广-从向量加法到-gemm-flashattention\">\u003Ca class=\"anchor\" href=\"#范式推广-从向量加法到-gemm-flashattention\" aria-label=\"跳转到标题：范式推广：从向量加法到 GEMM\u002FFlashAttention\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>范式推广：从向量加法到 GEMM\u002FFlashAttention\u003C\u002Fh3>\u003Cul>\n\u003Cli>万能公式：若干 Tensor → Arrangement 生成多组 Tiles → Application 逐组 Apply\u003C\u002Fli>\n\u003Cli>核心理念：记住这三行关系，即掌握了九齿的所有复杂算子编写逻辑\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw13.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=CehC%2FBTrUEB8JgP2IB134KkZHzk%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"双重可视化辅助理解\">\u003Ca class=\"anchor\" href=\"#双重可视化辅助理解\" aria-label=\"跳转到标题：双重可视化辅助理解\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>双重可视化辅助理解\u003C\u002Fh3>\u003Cul>\n\u003Cli>\u003Cstrong>嵌套张量可视化\u003C\u002Fstrong>：展示 Tile 后的多层结构（如长度 16 向量 → 外层 8×内层 2 的嵌套向量）\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw14.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=E42IWZUqouXl4R6hgJ9RV4MaBlQ%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cstrong>排布可视化\u003C\u002Fstrong>：交互展示每组 Tile 与原张量的位置对应关系\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_107_zw15.jpg?Expires=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Qf4IDMP4OB%2FSVdybmQH2AYn%2B79M%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>\u003Cem>完整代码示例与作业请访问课程仓库。如有疑问欢迎在 Issue 区交流讨论。\u003C\u002Fem>\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=1792750071&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=4vIkDijirVOEl01W4C%2B3fyE9q2Y%3D\" alt=\"\">\u003C\u002Fp>\n",[127,130,133,136,138,141,144,147,150,153,156,159,161,164,167,170,172,175,178,180,183,185],{"id":128,"text":129,"level":110},"一-领域特定语言-dsl","一、 领域特定语言（DSL）",{"id":131,"text":132,"level":9},"什么是-dsl","什么是 DSL？",{"id":134,"text":135,"level":9},"三者横向对比-以-向量加法-为例","三者横向对比：以“向量加法”为例",{"id":137,"text":137,"level":9},"设计出发点",{"id":139,"text":140,"level":110},"二-triton-编程基础","二、 Triton 编程基础",{"id":142,"text":143,"level":9},"为什么需要-triton","为什么需要 Triton？",{"id":145,"text":146,"level":9},"核心概念-从-thread-到-block","核心概念：从 Thread 到 Block",{"id":148,"text":149,"level":9},"四大核心-api","四大核心 API",{"id":151,"text":152,"level":110},"三-九齿-ninetoothed-核心设计思想","三、 九齿（NineToothed）核心设计思想",{"id":154,"text":155,"level":9},"九齿是什么","九齿是什么？",{"id":157,"text":158,"level":9},"为什么需要九齿","为什么需要九齿？",{"id":160,"text":160,"level":9},"核心范式",{"id":162,"text":163,"level":9},"代码对比-矩阵乘法-gemm","代码对比：矩阵乘法（GEMM）",{"id":165,"text":166,"level":110},"四-dsl-的实际用途","四、 DSL 的实际用途",{"id":168,"text":169,"level":9},"kernel-vs-operator-的概念区分","Kernel vs Operator 的概念区分",{"id":171,"text":171,"level":9},"核心应用场景",{"id":173,"text":174,"level":110},"五-向量加法","五、 向量加法",{"id":176,"text":177,"level":9},"triton-实现","Triton 实现",{"id":179,"text":179,"level":9},"九齿实现",{"id":181,"text":182,"level":9},"范式推广-从向量加法到-gemm-flashattention","范式推广：从向量加法到 GEMM\u002FFlashAttention",{"id":184,"text":184,"level":9},"双重可视化辅助理解",{"id":186,"text":186,"level":110},"关注与交流"]