[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"site-home-live-entry-setting":3,"site-header-contests":8,"site-header-camps":49,"study-material-detail-study-41":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_41_fm.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=%2BMR1MhyibXWdCwtakaQiGi6fFT8%3D","InfiniTensor,冬季训练营,Stable Diffusion,推理框架,工程实践","介绍 Stable Diffusion 在 InfiniTensor 推理框架中的适配流程，覆盖模型接入与工程化落地经验。","2026-10-22 20:38:11",2,"InfiniTensor 冬季训练营 AI 编译器第六课《Stable Diffusion 在 InfiniTensor 推理框架中的适配与工程实践》","![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_41_zw01.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=HLhGePrA8hnZnc2KUxrLkUL3Cz4%3D)\r\n\r\n2026 年 1 月 27 日，2025 冬季 InfiniTensor 大模型与人工智能系统训练营 AI 编译器系列第六课《Stable Diffusion 在 InfiniTensor 推理框架中的适配与工程实践》开讲。\r\n\r\n本节课介绍 Stable Diffusion 的基本概念、工作原理及其在 InfiniTensor 推理框架中的适配。\r\n\r\n\r\n## Stable Diffusion 介绍\r\n\r\n### Stable Diffusion 概述\r\n\r\n- **定义：** 基于扩散模型的生成式人工智能技术，主要用于文本生成图像，支持图像生成图像、局部重绘等多种任务。\r\n- **特点：** 通过潜在空间逐步去噪，将随机噪声还原为高质量图像，降低计算成本，支持在普通消费级 GPU 上运行。\r\n- **应用领域：** 艺术创作、游戏与影视、广告设计、科研探索等。\r\n\r\n### 扩散模型概念\r\n- **正向扩散：** 从有序到无序的固定加噪过程，由预定义的噪声调度控制。\r\n- **反向扩散：** 从无序到有序的去噪过程，模型学习如何逐步去除噪声。\r\n- **DDPM论文：** 介绍扩散模型的鼻祖，提及正向和反向扩散的步骤数。\r\n\r\n### Stable Diffusion 架构\r\n#### 1. 模块划分：\r\n- **文本语义编码：** 将自然语言描述转化为数值表示，建立文本与图像的语义对齐。\r\n- **迭代去噪生成：** 在潜在空间中逐步处理扩散信息，通过 U-NET+Scheduler 结构实现。\r\n- **空间重建解码：** 使用 VAE Decoder 将潜空间运算结果解码成实际图片。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_41_zw02.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=ZvmmcTw2rC80Q1RYDwHj60grssE%3D)\r\n\r\n#### 2. U-NET结构：\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_41_zw03.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=XhyGv5w%2Bd3PQON%2BWe2v0MJ3c2tg%3D)\r\n\r\n\r\n#### 3. DIT结构：\r\nDiT 采用纯 Transformer 结构作为去噪网络，与 Stable Diffusion 中常见的 U-Net 架构不同，代表了扩散模型在结构设计上的另一条技术路线。\r\n\r\n### 工作流\r\n#### 1. 文生图：\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_41_zw04.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=b2Fd6umxGuY%2FKz14Syiw%2F3ojym4%3D)\r\n\r\n\r\n#### 2. 图生图：\r\n- 图像输入：通过 VAE Encoder 将输入图像转化为潜变量。\r\n- 噪声注入：在特定时间步加入高斯噪声。\r\n- 迭代去噪：复用文生图的去噪过程，结合文本条件预测噪声。\r\n- 图像生成：通过 VAE Decoder 生成新图像。\r\n\r\n#### 3. 图像重绘：\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_41_zw05.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Kx1RgfliGsAHIr8y8T2k0bWJ3FM%3D)\r\n\r\n\r\n\r\n#### 4. 可控生成\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_41_zw06.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=QvsC2Mu7sc9%2FU4zXLycndLb1dNY%3D)\r\n\r\n- 控制图提取：从输入图像中提取边缘检测图、人体姿态图等控制图。\r\n- 条件注入：通过零卷积将控制图信息注入到 U-NET 结构中，实现语义和结构的双重控制。\r\n\r\n## Stable Diffusion 在 InfiniTensor 推理框架中的适配\r\n\r\n### 推理前准备\r\n#### 1. 环境搭建\r\n#### 2. 模型下载\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_41_zw07.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=0AQ5ntVCylNh71sp5s4slJu2BF8%3D)\r\n\r\n#### 3. 模型导出为 ONNX 文件\r\n#### 4. 模型预处理\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_41_zw08.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=rVm3jB75OUNKR0IzeJG7iLoJ15w%3D)\r\n\r\n\r\n### 推理过程\r\n- **1. 输入准备：** 创建批处理输入字典，指定prompt、batchsize、steps、guidance scales 等参数。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_41_zw09.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=7uPgB91Dr5FyPn3ZlbIp6aUR0nU%3D)\r\n\r\n- **2. 初始化常量参数：** 从 diffuser 库中获取潜在空间大小、是否做 CFG 操作、潜在空间通道维度等参数。\r\n- **3. 模型初始化：** 根据配置文件初始化文本编码器、U-NET\u002FDIT 模块、tokenizer、scheduler等。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_41_zw10.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=J95IP%2F1rzXHor7s2wPfNLiIuoL8%3D)\r\n\r\n- **4. 推理执行：**\r\n- 文本预处理：调用文本编码器（在部分模型配置下可能包含多个编码器）提取文本特征，并进行必要的特征对齐。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_41_zw11.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=ph2gTzY%2BZhiF%2Ft0%2BTjM1epKFH5w%3D)\r\n\r\n- 时间步生成：根据输入步骤数生成时间序列。\r\n- 潜在变量初始化：随机生成潜在空间变量。\r\n- 迭代去噪：根据时间步和文本嵌入，逐步去除噪声，更新潜在变量。\r\n- 图像生成：通过 VAE Decoder 生成最终图像，进行后处理（如颜色空间转化和归一化逆变换）。\r\n\r\n### 结果展示\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_41_zw12.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=vCzgYDIddJzbxN8aFgX6ZvR%2FByc%3D)\r\n\r\n\r\n\r\n## 总结\r\n本课详细讲述了 Stable Diffusion 的基本概念、工作原理及其在 InfiniTensor 推理框架中的适配过程。通过这节课让大家理解 Stable Diffusion 这一类的模型的整个工作流。\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-40","study-40","InfiniTensor 冬季训练营 CUDA 编程第六课《异步并行、底层控制与系统优化》",68,"\u002Fstudy-material\u002Fstudy-41","study-41","markdown",{"publishTime":7,"createTime":109,"canonicalPath":122,"id":123,"title":124},"\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》",{"htmlContent":126,"headings":127},"\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_41_zw01.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=HLhGePrA8hnZnc2KUxrLkUL3Cz4%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>2026 年 1 月 27 日，2025 冬季 InfiniTensor 大模型与人工智能系统训练营 AI 编译器系列第六课《Stable Diffusion 在 InfiniTensor 推理框架中的适配与工程实践》开讲。\u003C\u002Fp>\n\u003Cp>本节课介绍 Stable Diffusion 的基本概念、工作原理及其在 InfiniTensor 推理框架中的适配。\u003C\u002Fp>\n\u003Ch2 id=\"stable-diffusion-介绍\">\u003Ca class=\"anchor\" href=\"#stable-diffusion-介绍\" aria-label=\"跳转到标题：Stable Diffusion 介绍\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>Stable Diffusion 介绍\u003C\u002Fh2>\u003Ch3 id=\"stable-diffusion-概述\">\u003Ca class=\"anchor\" href=\"#stable-diffusion-概述\" aria-label=\"跳转到标题：Stable Diffusion 概述\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>Stable Diffusion 概述\u003C\u002Fh3>\u003Cul>\n\u003Cli>\u003Cstrong>定义：\u003C\u002Fstrong> 基于扩散模型的生成式人工智能技术，主要用于文本生成图像，支持图像生成图像、局部重绘等多种任务。\u003C\u002Fli>\n\u003Cli>\u003Cstrong>特点：\u003C\u002Fstrong> 通过潜在空间逐步去噪，将随机噪声还原为高质量图像，降低计算成本，支持在普通消费级 GPU 上运行。\u003C\u002Fli>\n\u003Cli>\u003Cstrong>应用领域：\u003C\u002Fstrong> 艺术创作、游戏与影视、广告设计、科研探索等。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch3 id=\"扩散模型概念\">\u003Ca class=\"anchor\" href=\"#扩散模型概念\" aria-label=\"跳转到标题：扩散模型概念\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>扩散模型概念\u003C\u002Fh3>\u003Cul>\n\u003Cli>\u003Cstrong>正向扩散：\u003C\u002Fstrong> 从有序到无序的固定加噪过程，由预定义的噪声调度控制。\u003C\u002Fli>\n\u003Cli>\u003Cstrong>反向扩散：\u003C\u002Fstrong> 从无序到有序的去噪过程，模型学习如何逐步去除噪声。\u003C\u002Fli>\n\u003Cli>\u003Cstrong>DDPM论文：\u003C\u002Fstrong> 介绍扩散模型的鼻祖，提及正向和反向扩散的步骤数。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch3 id=\"stable-diffusion-架构\">\u003Ca class=\"anchor\" href=\"#stable-diffusion-架构\" aria-label=\"跳转到标题：Stable Diffusion 架构\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>Stable Diffusion 架构\u003C\u002Fh3>\u003Ch4 id=\"1-模块划分\">\u003Ca class=\"anchor\" href=\"#1-模块划分\" aria-label=\"跳转到标题：1. 模块划分：\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>1. 模块划分：\u003C\u002Fh4>\u003Cul>\n\u003Cli>\u003Cstrong>文本语义编码：\u003C\u002Fstrong> 将自然语言描述转化为数值表示，建立文本与图像的语义对齐。\u003C\u002Fli>\n\u003Cli>\u003Cstrong>迭代去噪生成：\u003C\u002Fstrong> 在潜在空间中逐步处理扩散信息，通过 U-NET+Scheduler 结构实现。\u003C\u002Fli>\n\u003Cli>\u003Cstrong>空间重建解码：\u003C\u002Fstrong> 使用 VAE Decoder 将潜空间运算结果解码成实际图片。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_41_zw02.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=ZvmmcTw2rC80Q1RYDwHj60grssE%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"2-u-net结构\">\u003Ca class=\"anchor\" href=\"#2-u-net结构\" aria-label=\"跳转到标题：2. U-NET结构：\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>2. U-NET结构：\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_41_zw03.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=XhyGv5w%2Bd3PQON%2BWe2v0MJ3c2tg%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"3-dit结构\">\u003Ca class=\"anchor\" href=\"#3-dit结构\" aria-label=\"跳转到标题：3. DIT结构：\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>3. DIT结构：\u003C\u002Fh4>\u003Cp>DiT 采用纯 Transformer 结构作为去噪网络，与 Stable Diffusion 中常见的 U-Net 架构不同，代表了扩散模型在结构设计上的另一条技术路线。\u003C\u002Fp>\n\u003Ch3 id=\"工作流\">\u003Ca class=\"anchor\" href=\"#工作流\" aria-label=\"跳转到标题：工作流\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>工作流\u003C\u002Fh3>\u003Ch4 id=\"1-文生图\">\u003Ca class=\"anchor\" href=\"#1-文生图\" aria-label=\"跳转到标题：1. 文生图：\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>1. 文生图：\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_41_zw04.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=b2Fd6umxGuY%2FKz14Syiw%2F3ojym4%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"2-图生图\">\u003Ca class=\"anchor\" href=\"#2-图生图\" aria-label=\"跳转到标题：2. 图生图：\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>2. 图生图：\u003C\u002Fh4>\u003Cul>\n\u003Cli>图像输入：通过 VAE Encoder 将输入图像转化为潜变量。\u003C\u002Fli>\n\u003Cli>噪声注入：在特定时间步加入高斯噪声。\u003C\u002Fli>\n\u003Cli>迭代去噪：复用文生图的去噪过程，结合文本条件预测噪声。\u003C\u002Fli>\n\u003Cli>图像生成：通过 VAE Decoder 生成新图像。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch4 id=\"3-图像重绘\">\u003Ca class=\"anchor\" href=\"#3-图像重绘\" aria-label=\"跳转到标题：3. 图像重绘：\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>3. 图像重绘：\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_41_zw05.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Kx1RgfliGsAHIr8y8T2k0bWJ3FM%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"4-可控生成\">\u003Ca class=\"anchor\" href=\"#4-可控生成\" aria-label=\"跳转到标题：4. 可控生成\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>4. 可控生成\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_41_zw06.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=QvsC2Mu7sc9%2FU4zXLycndLb1dNY%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cul>\n\u003Cli>控制图提取：从输入图像中提取边缘检测图、人体姿态图等控制图。\u003C\u002Fli>\n\u003Cli>条件注入：通过零卷积将控制图信息注入到 U-NET 结构中，实现语义和结构的双重控制。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2 id=\"stable-diffusion-在-infinitensor-推理框架中的适配\">\u003Ca class=\"anchor\" href=\"#stable-diffusion-在-infinitensor-推理框架中的适配\" aria-label=\"跳转到标题：Stable Diffusion 在 InfiniTensor 推理框架中的适配\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>Stable Diffusion 在 InfiniTensor 推理框架中的适配\u003C\u002Fh2>\u003Ch3 id=\"推理前准备\">\u003Ca class=\"anchor\" href=\"#推理前准备\" aria-label=\"跳转到标题：推理前准备\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>推理前准备\u003C\u002Fh3>\u003Ch4 id=\"1-环境搭建\">\u003Ca class=\"anchor\" href=\"#1-环境搭建\" aria-label=\"跳转到标题：1. 环境搭建\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>1. 环境搭建\u003C\u002Fh4>\u003Ch4 id=\"2-模型下载\">\u003Ca class=\"anchor\" href=\"#2-模型下载\" aria-label=\"跳转到标题：2. 模型下载\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>2. 模型下载\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_41_zw07.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=0AQ5ntVCylNh71sp5s4slJu2BF8%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"3-模型导出为-onnx-文件\">\u003Ca class=\"anchor\" href=\"#3-模型导出为-onnx-文件\" aria-label=\"跳转到标题：3. 模型导出为 ONNX 文件\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>3. 模型导出为 ONNX 文件\u003C\u002Fh4>\u003Ch4 id=\"4-模型预处理\">\u003Ca class=\"anchor\" href=\"#4-模型预处理\" aria-label=\"跳转到标题：4. 模型预处理\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>4. 模型预处理\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_41_zw08.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=rVm3jB75OUNKR0IzeJG7iLoJ15w%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"推理过程\">\u003Ca class=\"anchor\" href=\"#推理过程\" aria-label=\"跳转到标题：推理过程\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>推理过程\u003C\u002Fh3>\u003Cul>\n\u003Cli>\u003Cstrong>1. 输入准备：\u003C\u002Fstrong> 创建批处理输入字典，指定prompt、batchsize、steps、guidance scales 等参数。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_41_zw09.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=7uPgB91Dr5FyPn3ZlbIp6aUR0nU%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cstrong>2. 初始化常量参数：\u003C\u002Fstrong> 从 diffuser 库中获取潜在空间大小、是否做 CFG 操作、潜在空间通道维度等参数。\u003C\u002Fli>\n\u003Cli>\u003Cstrong>3. 模型初始化：\u003C\u002Fstrong> 根据配置文件初始化文本编码器、U-NET\u002FDIT 模块、tokenizer、scheduler等。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_41_zw10.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=J95IP%2F1rzXHor7s2wPfNLiIuoL8%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cstrong>4. 推理执行：\u003C\u002Fstrong>\u003C\u002Fli>\n\u003Cli>文本预处理：调用文本编码器（在部分模型配置下可能包含多个编码器）提取文本特征，并进行必要的特征对齐。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_41_zw11.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=ph2gTzY%2BZhiF%2Ft0%2BTjM1epKFH5w%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cul>\n\u003Cli>时间步生成：根据输入步骤数生成时间序列。\u003C\u002Fli>\n\u003Cli>潜在变量初始化：随机生成潜在空间变量。\u003C\u002Fli>\n\u003Cli>迭代去噪：根据时间步和文本嵌入，逐步去除噪声，更新潜在变量。\u003C\u002Fli>\n\u003Cli>图像生成：通过 VAE Decoder 生成最终图像，进行后处理（如颜色空间转化和归一化逆变换）。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch3 id=\"结果展示\">\u003Ca class=\"anchor\" href=\"#结果展示\" aria-label=\"跳转到标题：结果展示\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>结果展示\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_41_zw12.jpg?Expires=1792752563&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=vCzgYDIddJzbxN8aFgX6ZvR%2FByc%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch2 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\u002Fh2>\u003Cp>本课详细讲述了 Stable Diffusion 的基本概念、工作原理及其在 InfiniTensor 推理框架中的适配过程。通过这节课让大家理解 Stable Diffusion 这一类的模型的整个工作流。\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",[128,131,134,136,139,143,146,149,151,154,157,160,163,166,168,171,174,177,180,182,184,186,189],{"id":129,"text":130,"level":110},"stable-diffusion-介绍","Stable Diffusion 介绍",{"id":132,"text":133,"level":9},"stable-diffusion-概述","Stable Diffusion 概述",{"id":135,"text":135,"level":9},"扩散模型概念",{"id":137,"text":138,"level":9},"stable-diffusion-架构","Stable Diffusion 架构",{"id":140,"text":141,"level":142},"1-模块划分","1. 模块划分：",4,{"id":144,"text":145,"level":142},"2-u-net结构","2. U-NET结构：",{"id":147,"text":148,"level":142},"3-dit结构","3. DIT结构：",{"id":150,"text":150,"level":9},"工作流",{"id":152,"text":153,"level":142},"1-文生图","1. 文生图：",{"id":155,"text":156,"level":142},"2-图生图","2. 图生图：",{"id":158,"text":159,"level":142},"3-图像重绘","3. 图像重绘：",{"id":161,"text":162,"level":142},"4-可控生成","4. 可控生成",{"id":164,"text":165,"level":110},"stable-diffusion-在-infinitensor-推理框架中的适配","Stable Diffusion 在 InfiniTensor 推理框架中的适配",{"id":167,"text":167,"level":9},"推理前准备",{"id":169,"text":170,"level":142},"1-环境搭建","1. 环境搭建",{"id":172,"text":173,"level":142},"2-模型下载","2. 模型下载",{"id":175,"text":176,"level":142},"3-模型导出为-onnx-文件","3. 模型导出为 ONNX 文件",{"id":178,"text":179,"level":142},"4-模型预处理","4. 模型预处理",{"id":181,"text":181,"level":9},"推理过程",{"id":183,"text":183,"level":9},"结果展示",{"id":185,"text":185,"level":110},"总结",{"id":187,"text":188,"level":9},"学习贴士","💡 学习贴士",{"id":190,"text":191,"level":9},"训练营报名持续进行中","📚 训练营报名持续进行中"]