[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"site-home-live-entry-setting":3,"site-header-contests":8,"site-header-camps":49,"study-material-detail-study-49":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=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=DdQTMKVo1nful94EsDeKYRm5DII%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=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=HWdVGs1NgNzaJvqpNa66tZqqOTo%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=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=5IITjeecr8z%2B54QsS2SaoytOVRM%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_49_fm.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=6iUp69cpjcVJ%2BNeS0GMACFD4fm4%3D","InfiniTensor,冬季训练营,vLLM-Ascend,大模型推理,性能优化","解读 vLLM-Ascend 推理方案，说明大模型推理架构、吞吐优化与工程化实践。","2026-10-22 20:38:11",2,"InfiniTensor 冬季训练营推理引擎第六课《vLLM-Ascend 大模型推理》","![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw01.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=xrAkXmuKQMb%2FAT0UeuDNJvHUBig%3D)\r\n\r\n\r\n2026 年 1 月 30 日，2025 冬季 InfiniTensor 大模型与人工智能系统训练营推理引擎系列第六课《vLLM-Ascend 大模型推理》开讲。\r\n\r\n本节课主要讲解 vLLM-ascend 整体规划和架构以及核心技术内容。\r\n\r\n\r\n\r\n## vLLM+vLLM-ascend 整体规划和架构\r\n\r\n### 社区演进概述\r\n- 回顾社区在推理引擎领域的发展历程。\r\n- 强调开源与性能并重的重要性。\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw02.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=snxIkMX5MznAcAaNkJvmgBPHYho%3D)\r\n\r\n\r\n### 华为昇腾技术路线\r\n- 以开源为基础，与社区紧密配合，回馈社区。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw03.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=ddOvJr7KmaltwDGpS4Qp49LpYLk%3D)\r\n\r\n\r\n\r\n### 昇腾推理架构解析\r\n#### 架构概览\r\n- 华为MindIE推理引擎与业界开源生态引擎（如LLM、TensorFlow TGI、Triton）的对比。\r\n#### 计算架构层\r\n- 对标 CUDA 的计算架构设计。\r\n#### 框架 Plugin 层\r\n- 介绍 vLLM-Ascend 作为 vLLM 框架的 Plugin，实现硬件层适配。\r\n\r\n### vLLM 核心技术\r\n#### 1. Paged Attention\r\n- 解决内存碎片与显存浪费问题，通过分页内存分配提高资源利用率。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw04.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=wIK46vtP6HVF9FqRbIQfRppEMcI%3D)\r\n\r\n#### 2. Continuous Batching\r\n- 与 Paged Attention 配套使用，提高 GPU\u002FNPU 算力和显存利用率。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw05.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=IeaTo%2FHHzvEuYhw9mHT8CYgiNdQ%3D)\r\n\r\n#### 3. vLLM-Ascend 插件\r\n- 硬件层适配，屏蔽 GPU 与 NPU 差异，支持无缝迁移。\r\n- 降低二次开发门槛，支持模型零拷贝。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw06.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=at9y4HQIY6TbQU3rSPgYXayv5%2FU%3D)\r\n\r\n#### 4. 安装与调用流程\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw07.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=uPrzh%2F8FWZ%2B3F5P4X3UgpPj98Bo%3D)\r\n\r\n\r\n#### 5. ACL Graph与性能优化\r\n##### （1）ACL Graph 概念\r\n- 对标CUDA graph，实现一次捕获、多次重放，减少空泡。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw08.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=6OidufD7GDI8IczT7VSia77t3TU%3D)\r\n\r\n##### （2）性能优化策略\r\n- 提高系统并行度，减少无法并行部分。\r\n- 介绍 vLLM 如何通过提高应用层系统并行度实现性能提升。\r\n\r\n## 推理精度分析方法\r\n### 精度分析工具\r\n1. 介绍 OpenCampass、EvalScope、AISBench 等工具。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw09.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=142Tm3RL9xFQQ9iq8sAbp1V2aCU%3D)\r\n\r\n2. 强调通过跑数据集获取精度指标的方法。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw10.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=K8T5he1spcEGwhiggbTqy7LlKm0%3D)\r\n\r\n\r\n### 常见问题与解决方案\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw11.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=k3CeW%2BjZnJrzfiFf0njZK%2FYa%2Fqc%3D)\r\n\r\n- 介绍如何通过调整参数和模型设计提高精度。\r\n\r\n### Badcase 分析\r\n1. 如何定位和分析 Badcase，找出精度不达标的原因。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw12.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=jZP4Z9iy%2FZvzyt3YsiZUZ89DVt8%3D)\r\n\r\n\r\n2. 基于MSProbe工具的精度数据采集\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw13.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=eT7Fo7aBESuzCD1iCxjph%2Ba9egM%3D)\r\n\r\n\r\n## 推理性能分析方法\r\n### 性能分析工具\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw14.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Vb08iSZBAh6wlFGJ7MaZLY9vwuM%3D)\r\n\r\n使用 AISBench 进行性能测评：\r\n- 配置随机数据集\r\n- 指定模型和数据集\r\n- 通过命令行启动性能测评\r\n\r\n### 性能问题定位\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw15.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=sMB%2FqNOnjwMUCyBR%2Fcj8BH%2Fu06A%3D)\r\n\r\n### 性能数据分析\r\n- `op_statistic.csv`\r\n- `trac_view.json`\r\n\r\n### PD 分离场景调优\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw16.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=c0zpw19KQnSg8z5gBn%2FsZZ76OA8%3D)\r\n\r\n- PD 分离场景调优：通过解耦预填充（Prefill）与生成（Decode）阶段，突破 KV Cache 导致的吞吐瓶颈。\r\n- 介绍如何通过调整 PD 配比和预期 TPS 来优化系统性能。\r\n\r\n## vLLM-ascend 推理部署流程\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw17.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=M%2Bt6BZL%2Fp%2FczVS%2Bj%2F9yOKIH8Csk%3D)\r\n\r\n### 环境准备\r\n- 使用 `npu-smi info` 检查 NPU 驱动固件安装情况。\r\n- 推荐使用 Docker 进行部署。\r\n\r\n### 模型下载与安装\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw18.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=h1HRTKXcSrBgeD0pEy1sz3Xr%2BAk%3D)\r\n\r\n- 介绍如何通过 ModelScope 下载模型。\r\n强调安装 vLLM 与 vLLM-Ascend 的步骤。\r\n\r\n### 离线与在线推理\r\n- 介绍离线与在线推理的使用方式与区别。\r\n- 强调 vLLM-Ascend 与 vLLM 使用方式的一致性。\r\n\r\n### 常用特性与入参介绍\r\n- **环境变量**\r\n- **推理参数**\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw19.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=8GdNnb6zP7aV0z1EUeIW0slOd1Q%3D)\r\n\r\n### 注意事项\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw20.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=TBzDw%2BOGKVCzIxZfGJ13MDDaj0c%3D)\r\n\r\n\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=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=tmWtTrH27PV8kQApQCBd1QaHaNU%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![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_ewm.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=O1NL5bahW5nvf%2FlqFbFObNY5fbM%3D)",{"publishTime":7,"createTime":109,"canonicalPath":114,"id":115,"title":116},"\u002Fstudy-material\u002Fstudy-48","study-48","InfiniTensor 冬季训练营 Triton&九齿第七课《九齿三重：天通》",60,"\u002Fstudy-material\u002Fstudy-49","study-49","markdown",{"publishTime":7,"createTime":109,"canonicalPath":122,"id":123,"title":124},"\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》",{"htmlContent":126,"headings":127},"\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw01.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=xrAkXmuKQMb%2FAT0UeuDNJvHUBig%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>2026 年 1 月 30 日，2025 冬季 InfiniTensor 大模型与人工智能系统训练营推理引擎系列第六课《vLLM-Ascend 大模型推理》开讲。\u003C\u002Fp>\n\u003Cp>本节课主要讲解 vLLM-ascend 整体规划和架构以及核心技术内容。\u003C\u002Fp>\n\u003Ch2 id=\"vllm-vllm-ascend-整体规划和架构\">\u003Ca class=\"anchor\" href=\"#vllm-vllm-ascend-整体规划和架构\" aria-label=\"跳转到标题：vLLM+vLLM-ascend 整体规划和架构\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>vLLM+vLLM-ascend 整体规划和架构\u003C\u002Fh2>\u003Ch3 id=\"社区演进概述\">\u003Ca class=\"anchor\" href=\"#社区演进概述\" aria-label=\"跳转到标题：社区演进概述\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>社区演进概述\u003C\u002Fh3>\u003Cul>\n\u003Cli>回顾社区在推理引擎领域的发展历程。\u003C\u002Fli>\n\u003Cli>强调开源与性能并重的重要性。\u003Cbr>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw02.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=snxIkMX5MznAcAaNkJvmgBPHYho%3D\" alt=\"\">\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch3 id=\"华为昇腾技术路线\">\u003Ca class=\"anchor\" href=\"#华为昇腾技术路线\" aria-label=\"跳转到标题：华为昇腾技术路线\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>华为昇腾技术路线\u003C\u002Fh3>\u003Cul>\n\u003Cli>以开源为基础，与社区紧密配合，回馈社区。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw03.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=ddOvJr7KmaltwDGpS4Qp49LpYLk%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"昇腾推理架构解析\">\u003Ca class=\"anchor\" href=\"#昇腾推理架构解析\" aria-label=\"跳转到标题：昇腾推理架构解析\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>昇腾推理架构解析\u003C\u002Fh3>\u003Ch4 id=\"架构概览\">\u003Ca class=\"anchor\" href=\"#架构概览\" aria-label=\"跳转到标题：架构概览\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>架构概览\u003C\u002Fh4>\u003Cul>\n\u003Cli>华为MindIE推理引擎与业界开源生态引擎（如LLM、TensorFlow TGI、Triton）的对比。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch4 id=\"计算架构层\">\u003Ca class=\"anchor\" href=\"#计算架构层\" aria-label=\"跳转到标题：计算架构层\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>计算架构层\u003C\u002Fh4>\u003Cul>\n\u003Cli>对标 CUDA 的计算架构设计。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch4 id=\"框架-plugin-层\">\u003Ca class=\"anchor\" href=\"#框架-plugin-层\" aria-label=\"跳转到标题：框架 Plugin 层\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>框架 Plugin 层\u003C\u002Fh4>\u003Cul>\n\u003Cli>介绍 vLLM-Ascend 作为 vLLM 框架的 Plugin，实现硬件层适配。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch3 id=\"vllm-核心技术\">\u003Ca class=\"anchor\" href=\"#vllm-核心技术\" aria-label=\"跳转到标题：vLLM 核心技术\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>vLLM 核心技术\u003C\u002Fh3>\u003Ch4 id=\"1-paged-attention\">\u003Ca class=\"anchor\" href=\"#1-paged-attention\" aria-label=\"跳转到标题：1. Paged Attention\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>1. Paged Attention\u003C\u002Fh4>\u003Cul>\n\u003Cli>解决内存碎片与显存浪费问题，通过分页内存分配提高资源利用率。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw04.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=wIK46vtP6HVF9FqRbIQfRppEMcI%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"2-continuous-batching\">\u003Ca class=\"anchor\" href=\"#2-continuous-batching\" aria-label=\"跳转到标题：2. Continuous Batching\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>2. Continuous Batching\u003C\u002Fh4>\u003Cul>\n\u003Cli>与 Paged Attention 配套使用，提高 GPU\u002FNPU 算力和显存利用率。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw05.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=IeaTo%2FHHzvEuYhw9mHT8CYgiNdQ%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"3-vllm-ascend-插件\">\u003Ca class=\"anchor\" href=\"#3-vllm-ascend-插件\" aria-label=\"跳转到标题：3. vLLM-Ascend 插件\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>3. vLLM-Ascend 插件\u003C\u002Fh4>\u003Cul>\n\u003Cli>硬件层适配，屏蔽 GPU 与 NPU 差异，支持无缝迁移。\u003C\u002Fli>\n\u003Cli>降低二次开发门槛，支持模型零拷贝。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw06.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=at9y4HQIY6TbQU3rSPgYXayv5%2FU%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"4-安装与调用流程\">\u003Ca class=\"anchor\" href=\"#4-安装与调用流程\" aria-label=\"跳转到标题：4. 安装与调用流程\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>4. 安装与调用流程\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw07.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=uPrzh%2F8FWZ%2B3F5P4X3UgpPj98Bo%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"5-acl-graph与性能优化\">\u003Ca class=\"anchor\" href=\"#5-acl-graph与性能优化\" aria-label=\"跳转到标题：5. ACL Graph与性能优化\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>5. ACL Graph与性能优化\u003C\u002Fh4>\u003Ch5 id=\"1-acl-graph-概念\">\u003Ca class=\"anchor\" href=\"#1-acl-graph-概念\" aria-label=\"跳转到标题：（1）ACL Graph 概念\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>（1）ACL Graph 概念\u003C\u002Fh5>\u003Cul>\n\u003Cli>对标CUDA graph，实现一次捕获、多次重放，减少空泡。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw08.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=6OidufD7GDI8IczT7VSia77t3TU%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch5 id=\"2-性能优化策略\">\u003Ca class=\"anchor\" href=\"#2-性能优化策略\" aria-label=\"跳转到标题：（2）性能优化策略\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>（2）性能优化策略\u003C\u002Fh5>\u003Cul>\n\u003Cli>提高系统并行度，减少无法并行部分。\u003C\u002Fli>\n\u003Cli>介绍 vLLM 如何通过提高应用层系统并行度实现性能提升。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2 id=\"推理精度分析方法\">\u003Ca class=\"anchor\" href=\"#推理精度分析方法\" aria-label=\"跳转到标题：推理精度分析方法\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>推理精度分析方法\u003C\u002Fh2>\u003Ch3 id=\"精度分析工具\">\u003Ca class=\"anchor\" href=\"#精度分析工具\" aria-label=\"跳转到标题：精度分析工具\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>精度分析工具\u003C\u002Fh3>\u003Col>\n\u003Cli>介绍 OpenCampass、EvalScope、AISBench 等工具。\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw09.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=142Tm3RL9xFQQ9iq8sAbp1V2aCU%3D\" alt=\"\">\u003C\u002Fp>\n\u003Col start=\"2\">\n\u003Cli>强调通过跑数据集获取精度指标的方法。\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw10.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=K8T5he1spcEGwhiggbTqy7LlKm0%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"常见问题与解决方案\">\u003Ca class=\"anchor\" href=\"#常见问题与解决方案\" aria-label=\"跳转到标题：常见问题与解决方案\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>常见问题与解决方案\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw11.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=k3CeW%2BjZnJrzfiFf0njZK%2FYa%2Fqc%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cul>\n\u003Cli>介绍如何通过调整参数和模型设计提高精度。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch3 id=\"badcase-分析\">\u003Ca class=\"anchor\" href=\"#badcase-分析\" aria-label=\"跳转到标题：Badcase 分析\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>Badcase 分析\u003C\u002Fh3>\u003Col>\n\u003Cli>如何定位和分析 Badcase，找出精度不达标的原因。\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw12.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=jZP4Z9iy%2FZvzyt3YsiZUZ89DVt8%3D\" alt=\"\">\u003C\u002Fp>\n\u003Col start=\"2\">\n\u003Cli>基于MSProbe工具的精度数据采集\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw13.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=eT7Fo7aBESuzCD1iCxjph%2Ba9egM%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch2 id=\"推理性能分析方法\">\u003Ca class=\"anchor\" href=\"#推理性能分析方法\" aria-label=\"跳转到标题：推理性能分析方法\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>推理性能分析方法\u003C\u002Fh2>\u003Ch3 id=\"性能分析工具\">\u003Ca class=\"anchor\" href=\"#性能分析工具\" aria-label=\"跳转到标题：性能分析工具\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>性能分析工具\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw14.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Vb08iSZBAh6wlFGJ7MaZLY9vwuM%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>使用 AISBench 进行性能测评：\u003C\u002Fp>\n\u003Cul>\n\u003Cli>配置随机数据集\u003C\u002Fli>\n\u003Cli>指定模型和数据集\u003C\u002Fli>\n\u003Cli>通过命令行启动性能测评\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch3 id=\"性能问题定位\">\u003Ca class=\"anchor\" href=\"#性能问题定位\" aria-label=\"跳转到标题：性能问题定位\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>性能问题定位\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw15.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=sMB%2FqNOnjwMUCyBR%2Fcj8BH%2Fu06A%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"性能数据分析\">\u003Ca class=\"anchor\" href=\"#性能数据分析\" aria-label=\"跳转到标题：性能数据分析\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>性能数据分析\u003C\u002Fh3>\u003Cul>\n\u003Cli>\u003Ccode>op_statistic.csv\u003C\u002Fcode>\u003C\u002Fli>\n\u003Cli>\u003Ccode>trac_view.json\u003C\u002Fcode>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch3 id=\"pd-分离场景调优\">\u003Ca class=\"anchor\" href=\"#pd-分离场景调优\" aria-label=\"跳转到标题：PD 分离场景调优\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>PD 分离场景调优\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw16.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=c0zpw19KQnSg8z5gBn%2FsZZ76OA8%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cul>\n\u003Cli>PD 分离场景调优：通过解耦预填充（Prefill）与生成（Decode）阶段，突破 KV Cache 导致的吞吐瓶颈。\u003C\u002Fli>\n\u003Cli>介绍如何通过调整 PD 配比和预期 TPS 来优化系统性能。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2 id=\"vllm-ascend-推理部署流程\">\u003Ca class=\"anchor\" href=\"#vllm-ascend-推理部署流程\" aria-label=\"跳转到标题：vLLM-ascend 推理部署流程\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>vLLM-ascend 推理部署流程\u003C\u002Fh2>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw17.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=M%2Bt6BZL%2Fp%2FczVS%2Bj%2F9yOKIH8Csk%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"环境准备\">\u003Ca class=\"anchor\" href=\"#环境准备\" aria-label=\"跳转到标题：环境准备\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>环境准备\u003C\u002Fh3>\u003Cul>\n\u003Cli>使用 \u003Ccode>npu-smi info\u003C\u002Fcode> 检查 NPU 驱动固件安装情况。\u003C\u002Fli>\n\u003Cli>推荐使用 Docker 进行部署。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch3 id=\"模型下载与安装\">\u003Ca class=\"anchor\" href=\"#模型下载与安装\" aria-label=\"跳转到标题：模型下载与安装\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>模型下载与安装\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw18.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=h1HRTKXcSrBgeD0pEy1sz3Xr%2BAk%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cul>\n\u003Cli>介绍如何通过 ModelScope 下载模型。\u003Cbr>强调安装 vLLM 与 vLLM-Ascend 的步骤。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch3 id=\"离线与在线推理\">\u003Ca class=\"anchor\" href=\"#离线与在线推理\" aria-label=\"跳转到标题：离线与在线推理\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>离线与在线推理\u003C\u002Fh3>\u003Cul>\n\u003Cli>介绍离线与在线推理的使用方式与区别。\u003C\u002Fli>\n\u003Cli>强调 vLLM-Ascend 与 vLLM 使用方式的一致性。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch3 id=\"常用特性与入参介绍\">\u003Ca class=\"anchor\" href=\"#常用特性与入参介绍\" aria-label=\"跳转到标题：常用特性与入参介绍\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>常用特性与入参介绍\u003C\u002Fh3>\u003Cul>\n\u003Cli>\u003Cstrong>环境变量\u003C\u002Fstrong>\u003C\u002Fli>\n\u003Cli>\u003Cstrong>推理参数\u003C\u002Fstrong>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw19.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=8GdNnb6zP7aV0z1EUeIW0slOd1Q%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"注意事项\">\u003Ca class=\"anchor\" href=\"#注意事项\" aria-label=\"跳转到标题：注意事项\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>注意事项\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_49_zw20.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=TBzDw%2BOGKVCzIxZfGJ13MDDaj0c%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_kcb.jpg?Expires=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=tmWtTrH27PV8kQApQCBd1QaHaNU%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=1792752825&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=O1NL5bahW5nvf%2FlqFbFObNY5fbM%3D\" alt=\"\">\u003C\u002Fp>\n",[128,131,133,135,137,140,142,145,148,151,154,157,160,163,166,169,171,173,175,178,180,182,184,186,189,192,194,196,198,200,202,205],{"id":129,"text":130,"level":110},"vllm-vllm-ascend-整体规划和架构","vLLM+vLLM-ascend 整体规划和架构",{"id":132,"text":132,"level":9},"社区演进概述",{"id":134,"text":134,"level":9},"华为昇腾技术路线",{"id":136,"text":136,"level":9},"昇腾推理架构解析",{"id":138,"text":138,"level":139},"架构概览",4,{"id":141,"text":141,"level":139},"计算架构层",{"id":143,"text":144,"level":139},"框架-plugin-层","框架 Plugin 层",{"id":146,"text":147,"level":9},"vllm-核心技术","vLLM 核心技术",{"id":149,"text":150,"level":139},"1-paged-attention","1. Paged Attention",{"id":152,"text":153,"level":139},"2-continuous-batching","2. Continuous Batching",{"id":155,"text":156,"level":139},"3-vllm-ascend-插件","3. vLLM-Ascend 插件",{"id":158,"text":159,"level":139},"4-安装与调用流程","4. 安装与调用流程",{"id":161,"text":162,"level":139},"5-acl-graph与性能优化","5. ACL Graph与性能优化",{"id":164,"text":165,"level":45},"1-acl-graph-概念","（1）ACL Graph 概念",{"id":167,"text":168,"level":45},"2-性能优化策略","（2）性能优化策略",{"id":170,"text":170,"level":110},"推理精度分析方法",{"id":172,"text":172,"level":9},"精度分析工具",{"id":174,"text":174,"level":9},"常见问题与解决方案",{"id":176,"text":177,"level":9},"badcase-分析","Badcase 分析",{"id":179,"text":179,"level":110},"推理性能分析方法",{"id":181,"text":181,"level":9},"性能分析工具",{"id":183,"text":183,"level":9},"性能问题定位",{"id":185,"text":185,"level":9},"性能数据分析",{"id":187,"text":188,"level":9},"pd-分离场景调优","PD 分离场景调优",{"id":190,"text":191,"level":110},"vllm-ascend-推理部署流程","vLLM-ascend 推理部署流程",{"id":193,"text":193,"level":9},"环境准备",{"id":195,"text":195,"level":9},"模型下载与安装",{"id":197,"text":197,"level":9},"离线与在线推理",{"id":199,"text":199,"level":9},"常用特性与入参介绍",{"id":201,"text":201,"level":9},"注意事项",{"id":203,"text":204,"level":9},"学习贴士","💡 学习贴士",{"id":206,"text":207,"level":9},"训练营报名持续进行中","📚 训练营报名持续进行中"]