[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"site-home-live-entry-setting":3,"site-header-contests":8,"site-header-camps":49,"study-material-detail-study-43":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=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Loo9mNcAiqrEJeoxKJGChNjsIP0%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=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=aQ7DEIPmeZ6G2GUnnO42fZDzkoY%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=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=rRUH7HCePjJ0x1Z1esoWYs0tjmE%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_43_fm.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=h%2F9ctP4uxqYBwZEnWp3hqoXxKRk%3D","InfiniTensor,冬季训练营,OpenCL,算子优化,性能调优","讲解 OpenCL 算子实现与优化方法，覆盖算子开发流程、性能分析与调优实践。","2026-10-22 20:38:11",2,"InfiniTensor 冬季训练营 OpenCL 编程第三课《OpenCL 算子的实现与优化》","![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw01.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Btdip%2BUcB8MaBMfeLxVF3AGq0Cc%3D)\r\n\r\n\r\n2026 年 1 月 28 日，2025 冬季 InfiniTensor 大模型与人工智能系统训练营 OpenCL 编程系列第三课《OpenCL 算子的实现与优化》开讲。\r\n\r\n本节课主要讲解 OpenCL 环境下常用算子的实现原理、性能瓶颈分析以及多种优化策略。\r\n\r\n## Argmax 算子实现\r\n- 1.Argmax 的定义与应用场景（如分类任务、大模型推理中的词选择）。\r\n- 2.单工作组实现方法（树状归约）与多工作组实现方法（分阶段求解，结合主机端同步）。\r\n- 3.数据划分、边界处理及线程任务分配原则。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw02.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=KHjlosvFNxVrY%2BQtUBsNvgloxO8%3D)\r\n\r\n\r\n## Softmax 算子实现\r\n### Softmax 的数学定义与改进公式\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw03.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=ZtLJ3nX9dPgzjg8FCMzimdiXfuc%3D)\r\n\r\n### 使用场景\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw04.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=PwY%2B9oAbObvPADQ6N0LzjtdTsuU%3D)\r\n\r\n\r\n\r\n\r\n## 矩阵乘法算子实现\r\n### 矩阵乘法定义及基础实现\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw05.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=fXTUbmPkEE9F3tXhw97zDZ9Oi2M%3D)\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw06.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=OaqQI4VAkhhaLWSOYXpg%2F4mQDV0%3D)\r\n\r\n### 优化方法\r\n#### 1. 子组优化\r\n- 子组内并行累加+子组归约，避免work-group级别的大量同步原子操作。\r\n- 能够合并内存访问，显著提高数据加载性能。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw07.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=sxQgXQh%2Bh6L2M%2FgPH%2FZO7XdKoZM%3D)\r\n\r\n\r\n\r\n#### 2. 向量化加载与计算\r\n- 使用向量类型（如 half8）和硬件友好尺寸。\r\n- 合并内存访问以提升带宽利用率。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw08.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=jIMSHUdtoMrydFVgNFK%2FpCjXrhU%3D)\r\n\r\n\r\n#### 3. 单工作项处理多数据\r\n- 根据数据规模调整工作项负载，减少调度开销。\r\n- 寄存器使用限制与边界处理。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw09.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=dEpRtkha%2BV0ZQdnjVxflo4IeluE%3D)\r\n\r\n\r\n## Gemv 量化\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw10.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=tt2cA%2FrWr4gDtE4ausVgOSnHwrI%3D)\r\n\r\n\r\n## GGUF 量化概述及量化实现\r\n### GGUF 量化概述\r\n#### 1. 传统量化方法\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw11.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Xnu4%2BT%2BLpZ4%2FJ7W3zhq6XWQ1ZlU%3D)\r\n\r\n\r\n#### 2. K 系列量化方法\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw12.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=TZyxyNsKFd70Vv%2BHCutXwDr%2FBbg%3D)\r\n\r\n#### 3.IQ 系列量化方法\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw13.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=iEaMc8IdRiMd2gaq3r15Fuak2d4%3D)\r\n\r\n\r\n### GGUF 量化实现\r\n#### 1. 实现思路\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw14.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=BE38AFLzyp8uL08lh58jc4aHN7A%3D)\r\n\r\n#### 2. 量化权重拆分\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw15.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=VIS%2BLhMjnmlrX5mAGIMpEutZLmA%3D)\r\n\r\n\r\n\r\n### 总结\r\n本节课讲解了 OpenCL 环境下常用算子（如Argmax、Softmax 及矩阵乘法）的实现原理、性能瓶颈分析以及多种优化策略，旨在提升 GPU 计算效率，特别针对大模型推理场景进行了特化优化。\r\n\r\n\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=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=UHjtW8jiWOaTgsS2PLaFF2a%2BW8A%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=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=v9ykgE0N%2B9nYRxtcFWbvLMcQFTY%3D)",{"publishTime":7,"createTime":109,"canonicalPath":114,"id":115,"title":116},"\u002Fstudy-material\u002Fstudy-42","study-42","InfiniTensor 冬季训练营 Triton & 九齿第六课《DLCompiler 的高性能算子开发实践》",66,"\u002Fstudy-material\u002Fstudy-43","study-43","markdown",{"publishTime":7,"createTime":109,"canonicalPath":122,"id":123,"title":124},"\u002Fstudy-material\u002Fstudy-44","study-44","InfiniTensor 冬季训练营 AI 编译器第七课《(MLIR)AscendNPU IR 编译堆栈》",{"htmlContent":126,"headings":127},"\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw01.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Btdip%2BUcB8MaBMfeLxVF3AGq0Cc%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>2026 年 1 月 28 日，2025 冬季 InfiniTensor 大模型与人工智能系统训练营 OpenCL 编程系列第三课《OpenCL 算子的实现与优化》开讲。\u003C\u002Fp>\n\u003Cp>本节课主要讲解 OpenCL 环境下常用算子的实现原理、性能瓶颈分析以及多种优化策略。\u003C\u002Fp>\n\u003Ch2 id=\"argmax-算子实现\">\u003Ca class=\"anchor\" href=\"#argmax-算子实现\" aria-label=\"跳转到标题：Argmax 算子实现\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>Argmax 算子实现\u003C\u002Fh2>\u003Cul>\n\u003Cli>1.Argmax 的定义与应用场景（如分类任务、大模型推理中的词选择）。\u003C\u002Fli>\n\u003Cli>2.单工作组实现方法（树状归约）与多工作组实现方法（分阶段求解，结合主机端同步）。\u003C\u002Fli>\n\u003Cli>3.数据划分、边界处理及线程任务分配原则。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw02.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=KHjlosvFNxVrY%2BQtUBsNvgloxO8%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch2 id=\"softmax-算子实现\">\u003Ca class=\"anchor\" href=\"#softmax-算子实现\" aria-label=\"跳转到标题：Softmax 算子实现\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>Softmax 算子实现\u003C\u002Fh2>\u003Ch3 id=\"softmax-的数学定义与改进公式\">\u003Ca class=\"anchor\" href=\"#softmax-的数学定义与改进公式\" aria-label=\"跳转到标题：Softmax 的数学定义与改进公式\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>Softmax 的数学定义与改进公式\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw03.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=ZtLJ3nX9dPgzjg8FCMzimdiXfuc%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"使用场景\">\u003Ca class=\"anchor\" href=\"#使用场景\" aria-label=\"跳转到标题：使用场景\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>使用场景\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw04.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=PwY%2B9oAbObvPADQ6N0LzjtdTsuU%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch2 id=\"矩阵乘法算子实现\">\u003Ca class=\"anchor\" href=\"#矩阵乘法算子实现\" aria-label=\"跳转到标题：矩阵乘法算子实现\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>矩阵乘法算子实现\u003C\u002Fh2>\u003Ch3 id=\"矩阵乘法定义及基础实现\">\u003Ca class=\"anchor\" href=\"#矩阵乘法定义及基础实现\" aria-label=\"跳转到标题：矩阵乘法定义及基础实现\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>矩阵乘法定义及基础实现\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw05.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=fXTUbmPkEE9F3tXhw97zDZ9Oi2M%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw06.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=OaqQI4VAkhhaLWSOYXpg%2F4mQDV0%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"优化方法\">\u003Ca class=\"anchor\" href=\"#优化方法\" aria-label=\"跳转到标题：优化方法\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>优化方法\u003C\u002Fh3>\u003Ch4 id=\"1-子组优化\">\u003Ca class=\"anchor\" href=\"#1-子组优化\" aria-label=\"跳转到标题：1. 子组优化\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>1. 子组优化\u003C\u002Fh4>\u003Cul>\n\u003Cli>子组内并行累加+子组归约，避免work-group级别的大量同步原子操作。\u003C\u002Fli>\n\u003Cli>能够合并内存访问，显著提高数据加载性能。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw07.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=sxQgXQh%2Bh6L2M%2FgPH%2FZO7XdKoZM%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"2-向量化加载与计算\">\u003Ca class=\"anchor\" href=\"#2-向量化加载与计算\" aria-label=\"跳转到标题：2. 向量化加载与计算\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>2. 向量化加载与计算\u003C\u002Fh4>\u003Cul>\n\u003Cli>使用向量类型（如 half8）和硬件友好尺寸。\u003C\u002Fli>\n\u003Cli>合并内存访问以提升带宽利用率。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw08.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=jIMSHUdtoMrydFVgNFK%2FpCjXrhU%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"3-单工作项处理多数据\">\u003Ca class=\"anchor\" href=\"#3-单工作项处理多数据\" aria-label=\"跳转到标题：3. 单工作项处理多数据\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>3. 单工作项处理多数据\u003C\u002Fh4>\u003Cul>\n\u003Cli>根据数据规模调整工作项负载，减少调度开销。\u003C\u002Fli>\n\u003Cli>寄存器使用限制与边界处理。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw09.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=dEpRtkha%2BV0ZQdnjVxflo4IeluE%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch2 id=\"gemv-量化\">\u003Ca class=\"anchor\" href=\"#gemv-量化\" aria-label=\"跳转到标题：Gemv 量化\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>Gemv 量化\u003C\u002Fh2>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw10.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=tt2cA%2FrWr4gDtE4ausVgOSnHwrI%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch2 id=\"gguf-量化概述及量化实现\">\u003Ca class=\"anchor\" href=\"#gguf-量化概述及量化实现\" aria-label=\"跳转到标题：GGUF 量化概述及量化实现\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>GGUF 量化概述及量化实现\u003C\u002Fh2>\u003Ch3 id=\"gguf-量化概述\">\u003Ca class=\"anchor\" href=\"#gguf-量化概述\" aria-label=\"跳转到标题：GGUF 量化概述\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>GGUF 量化概述\u003C\u002Fh3>\u003Ch4 id=\"1-传统量化方法\">\u003Ca class=\"anchor\" href=\"#1-传统量化方法\" aria-label=\"跳转到标题：1. 传统量化方法\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>1. 传统量化方法\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw11.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Xnu4%2BT%2BLpZ4%2FJ7W3zhq6XWQ1ZlU%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"2-k-系列量化方法\">\u003Ca class=\"anchor\" href=\"#2-k-系列量化方法\" aria-label=\"跳转到标题：2. K 系列量化方法\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>2. K 系列量化方法\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw12.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=TZyxyNsKFd70Vv%2BHCutXwDr%2FBbg%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"3-iq-系列量化方法\">\u003Ca class=\"anchor\" href=\"#3-iq-系列量化方法\" aria-label=\"跳转到标题：3.IQ 系列量化方法\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>3.IQ 系列量化方法\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw13.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=iEaMc8IdRiMd2gaq3r15Fuak2d4%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"gguf-量化实现\">\u003Ca class=\"anchor\" href=\"#gguf-量化实现\" aria-label=\"跳转到标题：GGUF 量化实现\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>GGUF 量化实现\u003C\u002Fh3>\u003Ch4 id=\"1-实现思路\">\u003Ca class=\"anchor\" href=\"#1-实现思路\" aria-label=\"跳转到标题：1. 实现思路\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>1. 实现思路\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw14.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=BE38AFLzyp8uL08lh58jc4aHN7A%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"2-量化权重拆分\">\u003Ca class=\"anchor\" href=\"#2-量化权重拆分\" aria-label=\"跳转到标题：2. 量化权重拆分\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>2. 量化权重拆分\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_43_zw15.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=VIS%2BLhMjnmlrX5mAGIMpEutZLmA%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"总结\">\u003Ca class=\"anchor\" href=\"#总结\" aria-label=\"跳转到标题：总结\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>总结\u003C\u002Fh3>\u003Cp>本节课讲解了 OpenCL 环境下常用算子（如Argmax、Softmax 及矩阵乘法）的实现原理、性能瓶颈分析以及多种优化策略，旨在提升 GPU 计算效率，特别针对大模型推理场景进行了特化优化。\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_kcb.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=UHjtW8jiWOaTgsS2PLaFF2a%2BW8A%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=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=v9ykgE0N%2B9nYRxtcFWbvLMcQFTY%3D\" alt=\"\">\u003C\u002Fp>\n",[128,131,134,137,139,141,143,145,149,152,155,158,161,164,167,170,173,176,179,182,184,187],{"id":129,"text":130,"level":110},"argmax-算子实现","Argmax 算子实现",{"id":132,"text":133,"level":110},"softmax-算子实现","Softmax 算子实现",{"id":135,"text":136,"level":9},"softmax-的数学定义与改进公式","Softmax 的数学定义与改进公式",{"id":138,"text":138,"level":9},"使用场景",{"id":140,"text":140,"level":110},"矩阵乘法算子实现",{"id":142,"text":142,"level":9},"矩阵乘法定义及基础实现",{"id":144,"text":144,"level":9},"优化方法",{"id":146,"text":147,"level":148},"1-子组优化","1. 子组优化",4,{"id":150,"text":151,"level":148},"2-向量化加载与计算","2. 向量化加载与计算",{"id":153,"text":154,"level":148},"3-单工作项处理多数据","3. 单工作项处理多数据",{"id":156,"text":157,"level":110},"gemv-量化","Gemv 量化",{"id":159,"text":160,"level":110},"gguf-量化概述及量化实现","GGUF 量化概述及量化实现",{"id":162,"text":163,"level":9},"gguf-量化概述","GGUF 量化概述",{"id":165,"text":166,"level":148},"1-传统量化方法","1. 传统量化方法",{"id":168,"text":169,"level":148},"2-k-系列量化方法","2. K 系列量化方法",{"id":171,"text":172,"level":148},"3-iq-系列量化方法","3.IQ 系列量化方法",{"id":174,"text":175,"level":9},"gguf-量化实现","GGUF 量化实现",{"id":177,"text":178,"level":148},"1-实现思路","1. 实现思路",{"id":180,"text":181,"level":148},"2-量化权重拆分","2. 量化权重拆分",{"id":183,"text":183,"level":9},"总结",{"id":185,"text":186,"level":9},"学习贴士","💡 学习贴士",{"id":188,"text":189,"level":9},"训练营报名持续进行中","📚 训练营报名持续进行中"]