[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"site-home-live-entry-setting":3,"site-header-contests":8,"site-header-camps":49,"study-material-detail-study-102":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=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=PPoBgu7QbKO%2B8J1uZA6NYzNftJM%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=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Tdn1aV9gvPHllFof%2FqUZvyavFPg%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=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=H4VCcV8qWm9Z2F5ZRtePaLur0Rg%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":47,"canonicalPath":117,"id":118,"contentType":119,"prevArticle":120},"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_fm.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=XbMCQqIrOByN7rCJQnBJjK8mLec%3D","多模态大模型,图文生成,Diffusion,LLaVA,多模态建模","2026 夏季 InfiniTensor 训练营《多模态智能及其应用》以 Nano Banana Pro 为落地案例拆解一体化多模态三大设计思路；梳理 CLIP \u002F SD 双技术演进线，讲解统一图文端到端架构；深度剖析解码、视觉特征、语义三层核心矛盾，对比四大建模派系优劣。","2026-10-22 20:38:11",2,"2026 夏季 InfiniTensor 训练营前沿模型 5：《多模态智能及其应用》","![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw01.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=IiJ%2Fbdi%2FhMyLHFh3YHUyW1x%2FuLc%3D)\r\n\r\n\u003Cspan style=\"font-size:12px;color:#555;\">8 月 12 日，2026 夏季 InfiniTensor 训练营前沿模型第五课《多模态智能及其应用》正式开讲。\r\n上一课《大模型自主智能体》讲解了单\u002F多智能体、工具学习、Agent 协作框架。\u003C\u002Fspan>\r\n\r\n本节课聚焦理解生成一体化多模态大模型，梳理其技术演进脉络、四大技术派系与工程落地案例 Nano Banana Pro。\r\n\r\n## Nano Banana Pro\r\n### 现象级案例解析\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw02.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=sNoRhjNAzAXKrsJRdRBJoydFKv4%3D)\r\n\r\n- ResNet 论文白板板书：英文拼写准确、框图逻辑严密\r\n- LLaVA-UHD v3 科普漫画：中文对白通俗、角色一致性强、文字渲染精准\r\n\r\n### Nano Banana Pro 的三大技术启示\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw03.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=%2FL7d8tpCPvyZn44nnj5uhMnnevU%3D)\r\n\r\n**1. 原生多模态框架**\r\n- 拒绝拼凑：底层模型必须统一在同一表征空间\r\n- 以 Gemini 类多模态基础模型为底座，同步学习理解与生成\r\n\r\n**2. 上下文多模态思考**\r\n- 超长上下文窗口支持：多图、多轮对话、长文本设定集\r\n- 从“一次性出图”进化为“对话式连续创作”\r\n\r\n**3. 高质量数据与评估体系**\r\n- 角色一致性来源：海量优质图文交织数据\r\n- 主观指标（美感\u002F还原度）：系统化人类评估体系驱动\r\n- 客观指标（文字\u002F布局）：高效量化反馈机制\r\n\r\n## 现有开源多模态生成模型\r\n### 现有模型的困境\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw04.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=65VfIsqkXPyuHJIQ7pBGwOMgigo%3D)\r\n\r\n### 理想 vs 现实：一体化框架的核心矛盾\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw05.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=8ceEdP%2B35Q8V7iQXrlUXCyjy8Tg%3D)\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw06.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=2tP904Kbhgk%2B%2Blzdk62qo%2B7fY6I%3D)\r\n\r\n**1. 编解码矛盾**：AR 适合文本但图像生成不直观；Diffusion 适合图像但难做文本理解\r\n**2. 视觉特征矛盾**：离散特征兼容 AR 但易退化；连续特征细节丰富但难与文本统一\r\n**3. 表示特性矛盾**：语义特征对齐好但缺细节；细节特征逼真但难支撑语义理解\r\n\r\n### 现有理解生成模型框架\r\n- **三维分析矩阵**：解码方式（AR\u002FDiffusion）× 视觉特征（连续\u002F离散）× 表示特性（语义\u002F细节）\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw07.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=6mogxKPSoug%2BH2qaBKkH6O1wMto%3D)\r\n\r\n#### 派系一：生成式外挂派 AD-QC-LS（伪）\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw08.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=i8%2BIR1fjVNrrqEop1%2FfemnswptE%3D)\r\n\r\n- **代表工作**：BLIP-3o, Qwen-Image\r\n- **核心思路**：LLM 生成 Condition → 注入现成 Diffusion 模型\r\n- **优劣**：快速具备能力；参数冗余、图文特征割裂\r\n\r\n#### 派系二：极致统一派\r\n**1. A-Q-L**（EMU, Chameleon）\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw09.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=e5%2F%2B6EoljEGd11HR5DCSkvhKVi8%3D)\r\n\r\n\r\n**2. A-C-L**（NextStep-1）：VAE 保留细节，但语义理解弱、AR 解码不直观\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw10.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=CbLf9j5Hqepip1lbNvKc%2Fgc6D24%3D)\r\n\r\n**3. A-Q-S**（VILA-U）：双任务优化统一特征，但 VQ 局限仍在\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw11.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=XH5e3DU8rv1WM73LAY4MXSYj9ek%3D)\r\n\r\n**4. D-Q-L**（LLaDA-V, MMaDA）：采用扩散式语言\u002F多模态建模范式；在生成过程中通过迭代去噪预测离散 token。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw12.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Oj6N%2BDlA0Ruz%2FjxSJ5qMemSCgpw%3D)\r\n\r\n#### 派系三：融合建模派（主流路线）\r\n**1. A-Q-LS**（TokenFlow）：CLIP+ViT 融合 VQ，适配 AR 但受限于离散化\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw13.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=v7T7GuXupNL4EDee0yHheAo%2FauI%3D)\r\n\r\n**2. A-QC-SL**（Janus \u002F BAGEL）：特征解耦（独立 Expert + 共享 Attention），双 SOTA 但内生融合不足\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw14.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=roReyLKHAtWRL6OFNEIvBUGccZQ%3D)\r\n\r\n**3. AD-C-L**（Transfusion）：同一 Backbone 内 AR(文) + Diffusion(图)\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw15.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=zW943mIK28WjPdn1TPIZveSydrg%3D)\r\n\r\n**4. AD-Q-SL**（Show-o）：CLIP-ViT + VAE 融合后 VQ 离散化 + Diffusion 解码，兼顾语义细节但仍受 VQ 限制\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw16.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=8%2BA5TnMkCrl9EqNs2f3wUkzHOMY%3D)\r\n\r\n**5. AD-QC-SL**（Manzano、X-Omni、TAR）：全维度融合（CLIP ViT 离散词表 + 外挂 Diffusion），能力强但结构冗余\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw17.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=NVfgshOOlKAhqbXDYwhR5ZqaY4E%3D)\r\n\r\n**6. AD-C-SL**（Show-o2、Tuna）：连续特征 + 语义细节融合 + VAE Decoder + Flow Matching，在论文报告的多项基准上取得较强性能\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw18.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=5b%2F07OH86oBI5x7nqxY4EKP3Hpw%3D)\r\n\r\n### DeepMind 启发\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw19.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=2aLNdmznqqN57fM1BW%2BnMXm9LKI%3D)\r\n\r\n从公开研究成果来看，连续视觉表示与 Diffusion\u002FFlow Matching 是统一多模态生成的重要技术路线；\r\n\r\n## Nano Banana Pro 技术方案预判\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw20.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=sx3RRhAE1jkdhFjJnnr6x%2B823Aw%3D)\r\n\r\n以下内容为基于公开信息和现有研究路线的技术推测。\r\n\r\n\r\n\r\n\r\n\r\n## 加入训练营\r\n📺**完整课程内容，请观看直播或查看课程回放：**\r\n\r\n**InfiniTensor 官网**： https:\u002F\u002Fwww.infinitensor.com\r\n\r\n**B站直播**： InfiniTensor 官方直播间\r\n\r\n**视频号直播**： InfiniTensor 视频号预约\r\n\r\n**答疑交流**： 训练营官方社群⬇️\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002Fqun_1072.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=07ovyfEz93I1Glk%2BjPdOxVw3ra4%3D)\r\n\r\n## 关注与交流\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002Fportal_news\u002Fnews_ewm02.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=yQQCxNoB64Iue0OdGDsKbeJjW00%3D)",{"publishTime":7,"createTime":109,"canonicalPath":114,"id":115,"title":116},"\u002Fstudy-material\u002Fstudy-101","study-101","2026 夏季 InfiniTensor 训练营推理引擎 6：《InfiniLM》","\u002Fstudy-material\u002Fstudy-102","study-102","markdown",{"publishTime":7,"createTime":109,"canonicalPath":121,"id":122,"title":123},"\u002Fstudy-material\u002Fstudy-103","study-103","2026 夏季 InfiniTensor 训练营通信与并行 3、推理引擎 7、训练系统 6：《基于九源平台的国产化集群应用实践及优化》",{"htmlContent":125,"headings":126},"\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw01.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=IiJ%2Fbdi%2FhMyLHFh3YHUyW1x%2FuLc%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>\u003Cspan style=\"font-size:12px;color:#555;\">8 月 12 日，2026 夏季 InfiniTensor 训练营前沿模型第五课《多模态智能及其应用》正式开讲。\u003Cbr>上一课《大模型自主智能体》讲解了单\u002F多智能体、工具学习、Agent 协作框架。\u003C\u002Fspan>\u003C\u002Fp>\n\u003Cp>本节课聚焦理解生成一体化多模态大模型，梳理其技术演进脉络、四大技术派系与工程落地案例 Nano Banana Pro。\u003C\u002Fp>\n\u003Ch2 id=\"nano-banana-pro\">\u003Ca class=\"anchor\" href=\"#nano-banana-pro\" aria-label=\"跳转到标题：Nano Banana Pro\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>Nano Banana Pro\u003C\u002Fh2>\u003Ch3 id=\"现象级案例解析\">\u003Ca class=\"anchor\" href=\"#现象级案例解析\" aria-label=\"跳转到标题：现象级案例解析\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>现象级案例解析\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw02.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=sNoRhjNAzAXKrsJRdRBJoydFKv4%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cul>\n\u003Cli>ResNet 论文白板板书：英文拼写准确、框图逻辑严密\u003C\u002Fli>\n\u003Cli>LLaVA-UHD v3 科普漫画：中文对白通俗、角色一致性强、文字渲染精准\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch3 id=\"nano-banana-pro-的三大技术启示\">\u003Ca class=\"anchor\" href=\"#nano-banana-pro-的三大技术启示\" aria-label=\"跳转到标题：Nano Banana Pro 的三大技术启示\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>Nano Banana Pro 的三大技术启示\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw03.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=%2FL7d8tpCPvyZn44nnj5uhMnnevU%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>\u003Cstrong>1. 原生多模态框架\u003C\u002Fstrong>\u003C\u002Fp>\n\u003Cul>\n\u003Cli>拒绝拼凑：底层模型必须统一在同一表征空间\u003C\u002Fli>\n\u003Cli>以 Gemini 类多模态基础模型为底座，同步学习理解与生成\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cstrong>2. 上下文多模态思考\u003C\u002Fstrong>\u003C\u002Fp>\n\u003Cul>\n\u003Cli>超长上下文窗口支持：多图、多轮对话、长文本设定集\u003C\u002Fli>\n\u003Cli>从“一次性出图”进化为“对话式连续创作”\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cstrong>3. 高质量数据与评估体系\u003C\u002Fstrong>\u003C\u002Fp>\n\u003Cul>\n\u003Cli>角色一致性来源：海量优质图文交织数据\u003C\u002Fli>\n\u003Cli>主观指标（美感\u002F还原度）：系统化人类评估体系驱动\u003C\u002Fli>\n\u003Cli>客观指标（文字\u002F布局）：高效量化反馈机制\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2 id=\"现有开源多模态生成模型\">\u003Ca class=\"anchor\" href=\"#现有开源多模态生成模型\" aria-label=\"跳转到标题：现有开源多模态生成模型\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>现有开源多模态生成模型\u003C\u002Fh2>\u003Ch3 id=\"现有模型的困境\">\u003Ca class=\"anchor\" href=\"#现有模型的困境\" aria-label=\"跳转到标题：现有模型的困境\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>现有模型的困境\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw04.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=65VfIsqkXPyuHJIQ7pBGwOMgigo%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"理想-vs-现实-一体化框架的核心矛盾\">\u003Ca class=\"anchor\" href=\"#理想-vs-现实-一体化框架的核心矛盾\" aria-label=\"跳转到标题：理想 vs 现实：一体化框架的核心矛盾\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>理想 vs 现实：一体化框架的核心矛盾\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw05.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=8ceEdP%2B35Q8V7iQXrlUXCyjy8Tg%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw06.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=2tP904Kbhgk%2B%2Blzdk62qo%2B7fY6I%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>\u003Cstrong>1. 编解码矛盾\u003C\u002Fstrong>：AR 适合文本但图像生成不直观；Diffusion 适合图像但难做文本理解\u003Cbr>\u003Cstrong>2. 视觉特征矛盾\u003C\u002Fstrong>：离散特征兼容 AR 但易退化；连续特征细节丰富但难与文本统一\u003Cbr>\u003Cstrong>3. 表示特性矛盾\u003C\u002Fstrong>：语义特征对齐好但缺细节；细节特征逼真但难支撑语义理解\u003C\u002Fp>\n\u003Ch3 id=\"现有理解生成模型框架\">\u003Ca class=\"anchor\" href=\"#现有理解生成模型框架\" aria-label=\"跳转到标题：现有理解生成模型框架\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>现有理解生成模型框架\u003C\u002Fh3>\u003Cul>\n\u003Cli>\u003Cstrong>三维分析矩阵\u003C\u002Fstrong>：解码方式（AR\u002FDiffusion）× 视觉特征（连续\u002F离散）× 表示特性（语义\u002F细节）\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw07.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=6mogxKPSoug%2BH2qaBKkH6O1wMto%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"派系一-生成式外挂派-ad-qc-ls-伪\">\u003Ca class=\"anchor\" href=\"#派系一-生成式外挂派-ad-qc-ls-伪\" aria-label=\"跳转到标题：派系一：生成式外挂派 AD-QC-LS（伪）\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>派系一：生成式外挂派 AD-QC-LS（伪）\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw08.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=i8%2BIR1fjVNrrqEop1%2FfemnswptE%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cstrong>代表工作\u003C\u002Fstrong>：BLIP-3o, Qwen-Image\u003C\u002Fli>\n\u003Cli>\u003Cstrong>核心思路\u003C\u002Fstrong>：LLM 生成 Condition → 注入现成 Diffusion 模型\u003C\u002Fli>\n\u003Cli>\u003Cstrong>优劣\u003C\u002Fstrong>：快速具备能力；参数冗余、图文特征割裂\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch4 id=\"派系二-极致统一派\">\u003Ca class=\"anchor\" href=\"#派系二-极致统一派\" aria-label=\"跳转到标题：派系二：极致统一派\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>派系二：极致统一派\u003C\u002Fh4>\u003Cp>\u003Cstrong>1. A-Q-L\u003C\u002Fstrong>（EMU, Chameleon）\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw09.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=e5%2F%2B6EoljEGd11HR5DCSkvhKVi8%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>\u003Cstrong>2. A-C-L\u003C\u002Fstrong>（NextStep-1）：VAE 保留细节，但语义理解弱、AR 解码不直观\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw10.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=CbLf9j5Hqepip1lbNvKc%2Fgc6D24%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>\u003Cstrong>3. A-Q-S\u003C\u002Fstrong>（VILA-U）：双任务优化统一特征，但 VQ 局限仍在\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw11.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=XH5e3DU8rv1WM73LAY4MXSYj9ek%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>\u003Cstrong>4. D-Q-L\u003C\u002Fstrong>（LLaDA-V, MMaDA）：采用扩散式语言\u002F多模态建模范式；在生成过程中通过迭代去噪预测离散 token。\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw12.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Oj6N%2BDlA0Ruz%2FjxSJ5qMemSCgpw%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"派系三-融合建模派-主流路线\">\u003Ca class=\"anchor\" href=\"#派系三-融合建模派-主流路线\" aria-label=\"跳转到标题：派系三：融合建模派（主流路线）\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>派系三：融合建模派（主流路线）\u003C\u002Fh4>\u003Cp>\u003Cstrong>1. A-Q-LS\u003C\u002Fstrong>（TokenFlow）：CLIP+ViT 融合 VQ，适配 AR 但受限于离散化\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw13.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=v7T7GuXupNL4EDee0yHheAo%2FauI%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>\u003Cstrong>2. A-QC-SL\u003C\u002Fstrong>（Janus \u002F BAGEL）：特征解耦（独立 Expert + 共享 Attention），双 SOTA 但内生融合不足\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw14.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=roReyLKHAtWRL6OFNEIvBUGccZQ%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>\u003Cstrong>3. AD-C-L\u003C\u002Fstrong>（Transfusion）：同一 Backbone 内 AR(文) + Diffusion(图)\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw15.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=zW943mIK28WjPdn1TPIZveSydrg%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>\u003Cstrong>4. AD-Q-SL\u003C\u002Fstrong>（Show-o）：CLIP-ViT + VAE 融合后 VQ 离散化 + Diffusion 解码，兼顾语义细节但仍受 VQ 限制\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw16.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=8%2BA5TnMkCrl9EqNs2f3wUkzHOMY%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>\u003Cstrong>5. AD-QC-SL\u003C\u002Fstrong>（Manzano、X-Omni、TAR）：全维度融合（CLIP ViT 离散词表 + 外挂 Diffusion），能力强但结构冗余\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw17.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=NVfgshOOlKAhqbXDYwhR5ZqaY4E%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>\u003Cstrong>6. AD-C-SL\u003C\u002Fstrong>（Show-o2、Tuna）：连续特征 + 语义细节融合 + VAE Decoder + Flow Matching，在论文报告的多项基准上取得较强性能\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw18.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=5b%2F07OH86oBI5x7nqxY4EKP3Hpw%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"deepmind-启发\">\u003Ca class=\"anchor\" href=\"#deepmind-启发\" aria-label=\"跳转到标题：DeepMind 启发\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>DeepMind 启发\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw19.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=2aLNdmznqqN57fM1BW%2BnMXm9LKI%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>从公开研究成果来看，连续视觉表示与 Diffusion\u002FFlow Matching 是统一多模态生成的重要技术路线；\u003C\u002Fp>\n\u003Ch2 id=\"nano-banana-pro-技术方案预判\">\u003Ca class=\"anchor\" href=\"#nano-banana-pro-技术方案预判\" aria-label=\"跳转到标题：Nano Banana Pro 技术方案预判\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>Nano Banana Pro 技术方案预判\u003C\u002Fh2>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_102_zw20.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=sx3RRhAE1jkdhFjJnnr6x%2B823Aw%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>以下内容为基于公开信息和现有研究路线的技术推测。\u003C\u002Fp>\n\u003Ch2 id=\"加入训练营\">\u003Ca class=\"anchor\" href=\"#加入训练营\" aria-label=\"跳转到标题：加入训练营\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>加入训练营\u003C\u002Fh2>\u003Cp>📺\u003Cstrong>完整课程内容，请观看直播或查看课程回放：\u003C\u002Fstrong>\u003C\u002Fp>\n\u003Cp>\u003Cstrong>InfiniTensor 官网\u003C\u002Fstrong>： \u003Ca href=\"https:\u002F\u002Fwww.infinitensor.com\">https:\u002F\u002Fwww.infinitensor.com\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp>\u003Cstrong>B站直播\u003C\u002Fstrong>： InfiniTensor 官方直播间\u003C\u002Fp>\n\u003Cp>\u003Cstrong>视频号直播\u003C\u002Fstrong>： InfiniTensor 视频号预约\u003C\u002Fp>\n\u003Cp>\u003Cstrong>答疑交流\u003C\u002Fstrong>： 训练营官方社群⬇️\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002Fqun_1072.jpg?Expires=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=07ovyfEz93I1Glk%2BjPdOxVw3ra4%3D\" alt=\"\">\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=1792750030&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=yQQCxNoB64Iue0OdGDsKbeJjW00%3D\" alt=\"\">\u003C\u002Fp>\n",[127,130,132,135,137,139,142,144,148,151,154,157,160,162],{"id":128,"text":129,"level":110},"nano-banana-pro","Nano Banana Pro",{"id":131,"text":131,"level":9},"现象级案例解析",{"id":133,"text":134,"level":9},"nano-banana-pro-的三大技术启示","Nano Banana Pro 的三大技术启示",{"id":136,"text":136,"level":110},"现有开源多模态生成模型",{"id":138,"text":138,"level":9},"现有模型的困境",{"id":140,"text":141,"level":9},"理想-vs-现实-一体化框架的核心矛盾","理想 vs 现实：一体化框架的核心矛盾",{"id":143,"text":143,"level":9},"现有理解生成模型框架",{"id":145,"text":146,"level":147},"派系一-生成式外挂派-ad-qc-ls-伪","派系一：生成式外挂派 AD-QC-LS（伪）",4,{"id":149,"text":150,"level":147},"派系二-极致统一派","派系二：极致统一派",{"id":152,"text":153,"level":147},"派系三-融合建模派-主流路线","派系三：融合建模派（主流路线）",{"id":155,"text":156,"level":9},"deepmind-启发","DeepMind 启发",{"id":158,"text":159,"level":110},"nano-banana-pro-技术方案预判","Nano Banana Pro 技术方案预判",{"id":161,"text":161,"level":110},"加入训练营",{"id":163,"text":163,"level":110},"关注与交流"]