[{"data":1,"prerenderedAt":150},["ShallowReactive",2],{"content-query-ehmJKwKI4d":3},{"_path":4,"_dir":5,"_draft":6,"_partial":6,"_locale":7,"title":8,"description":9,"date":10,"cover":11,"type":12,"state":13,"body":14,"_type":144,"_id":145,"_source":146,"_file":147,"_stem":148,"_extension":149},"/activities/zh/3643","zh",false,"","昇思框架×昇腾开发板：DeepSeek蒸馏版与多模态模型部署全流程实践","以"用昇思框架玩转DeepSeek开发实战" 为主题的昇腾开发板集训营火热开课。","2025-03-12","https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/03/14/9cda62b0fbd944d2bc30f7af4c209915.png","activities",1,{"type":15,"children":16,"toc":141},"root",[17,25,31,39,44,49,62,71,78,83,90,95,100,105,110,117,124,129,136],{"type":18,"tag":19,"props":20,"children":22},"element","h1",{"id":21},"昇思框架昇腾开发板deepseek蒸馏版与多模态模型部署全流程实践",[23],{"type":24,"value":8},"text",{"type":18,"tag":26,"props":27,"children":28},"p",{},[29],{"type":24,"value":30},"3月7日，在昇腾AI开发者创享日·浙江站分论坛上，以\"用昇思框架玩转DeepSeek开发实战\" 为主题的昇腾开发板集训营火热开课。活动吸引了来自高校师生与企业技术骨干的550余名开发者踊跃报名，最终70余名开发者通过筛选获得现场参与资格，共同开启昇腾开发板深度实践之旅。",{"type":18,"tag":26,"props":32,"children":33},{},[34],{"type":18,"tag":35,"props":36,"children":38},"img",{"alt":7,"src":37},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/03/14/c01afe4554b94d86988f2b004808c7b1.png",[],{"type":18,"tag":26,"props":40,"children":41},{},[42],{"type":24,"value":43},"培训中，昇思布道师非常work老师、Selina老师和香橙派系统工程师徐洋帆老师以深入浅出、详实风趣的讲演，吸引了现场每一位开发者。活动通过技术解析、模型适配实战与开发实操相结合的形式，深度展示了昇腾AI技术栈在边缘计算场景的创新应用。在深入体验香橙派开发板之后，大家纷纷表示不仅对开发有了更透彻的领悟，还燃起了创作创新的热情。",{"type":18,"tag":26,"props":45,"children":46},{},[47],{"type":24,"value":48},"本场昇腾开发板集训营系统讲解了昇腾AI基础软硬件和昇思MindSpore架构设计，深入剖析了Janus-Pro模型的结构特征、演进路径及推理机制，重点阐述了基于香橙派硬件平台与MindSpore框架的模型适配方案，通过算子适配、内存管理、性能优化等关键技术环节的实践验证，构建起端侧高效推理解决方案。",{"type":18,"tag":26,"props":50,"children":51},{},[52,54,60],{"type":24,"value":53},"昇思MindSpore布道师对MindSpore在香橙派开发板上的能力支持做了详细的说明，目前昇思MindSpore AI框架已实现在香橙派AIpro开发板预置，并在官方网站上线开发指导教程，通过软硬结合，使能开箱即用模型，并",{"type":18,"tag":55,"props":56,"children":57},"strong",{},[58],{"type":24,"value":59},"提供模型全流程",{"type":24,"value":61},"使能套件，实现积木式开发。不仅支持CV、NLP等小模型，同时满足小规格大模型（如DeepSeek-R1-Distill-Qwen-1.5B、DeepSeek-Janus-Pro-1B等）。开发过程中，动态图+类HuggingFace的编程方式给开发者提供更为灵活友好的模型开发与调试体验，而通过三行代码即可实现jit即时编译，获得更有的执行性能。",{"type":18,"tag":26,"props":63,"children":64},{},[65,69],{"type":18,"tag":35,"props":66,"children":68},{"alt":7,"src":67},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/03/14/24f75627a78f4632aca0994564e43fe9.png",[],{"type":24,"value":70}," 同时，活动有幸邀请到了香橙派系统工程师徐洋帆老师介绍了香橙派AIpro和AI Studio开发板的设计、能力和相关资源。",{"type":18,"tag":26,"props":72,"children":73},{},[74],{"type":18,"tag":35,"props":75,"children":77},{"alt":7,"src":76},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/03/14/6fd5f9412fd4458ca7141a4fcd78e1f1.png",[],{"type":18,"tag":26,"props":79,"children":80},{},[81],{"type":24,"value":82},"最后设置的全流程实践环节，以香橙派为载体，通过环境配置、模型下载、推理验证等实操步骤，让开发者们沉浸式体验了基于MindSpore的DeepSeek Janus Pro模型在香橙派AIpro开发板上的完整部署流程。",{"type":18,"tag":26,"props":84,"children":85},{},[86],{"type":18,"tag":35,"props":87,"children":89},{"alt":7,"src":88},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/03/14/16941dfc8ff2407892fddf7fc5c5ec13.png",[],{"type":18,"tag":26,"props":91,"children":92},{},[93],{"type":24,"value":94},"培训内容分为三个主要部分，讲师们首先从DeepSeek的Janus-Pro模型介绍、MindSpore框架在香橙派上的适配问题和解决方案进行介绍，并对开发者在开发板上的部署和推理过程进行悉心指导。",{"type":18,"tag":26,"props":96,"children":97},{},[98],{"type":24,"value":99},"首先，在Janus-Pro模型介绍部分，培训重点介绍了模型的双任务统一（多模态理解和生成），以及其架构上的创新。Janus-Pro通过分离理解和生成的编码器，解决了传统方法中的问题，并利用LLaMa结构实现统一，提升效率和复用性。",{"type":18,"tag":26,"props":101,"children":102},{},[103],{"type":24,"value":104},"第二部分在讲解MindSpore框架适配时，谈到接口转换、调试和性能优化等具体事项。在接口转换方面，讲师们提示，MindSpore NLP套件提供了兼容方案，用于处理MindSpore与PyTorch的接口差异。调试时建议使用动态图和同步模式，处理算子缺失问题。",{"type":18,"tag":26,"props":106,"children":107},{},[108],{"type":24,"value":109},"第三部分在介绍部署流程时，培训提到系统烧录的注意事项，环境变量的重要性，以及在香橙派上运行时的特殊考虑，如FP16支持和swap设置等。同时，讲师们提醒开发者根据指导手册进行操作，并注意常见问题的解决方法。",{"type":18,"tag":26,"props":111,"children":112},{},[113],{"type":18,"tag":35,"props":114,"children":116},{"alt":7,"src":115},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/03/14/4fce310ee6c54711bd042a78ac479cfc.png",[],{"type":18,"tag":26,"props":118,"children":119},{},[120],{"type":18,"tag":35,"props":121,"children":123},{"alt":7,"src":122},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/03/14/502ed92a06224797bd378108d4f0335b.png",[],{"type":18,"tag":26,"props":125,"children":126},{},[127],{"type":24,"value":128},"在其后的QA互动环节，开发者围绕模型部署与性能优化等问题展开了针对性提问。讲师们通过直面端侧部署痛点，通过内存管理、性能分析与集群扩展等实践问题的讲解，为开发者提供了可落地的优化思路与资源渠道，强化了昇思MindSpore在端侧场景的工程实践能力。",{"type":18,"tag":26,"props":130,"children":131},{},[132],{"type":18,"tag":35,"props":133,"children":135},{"alt":7,"src":134},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/03/14/42bd1469b60f45b3bf6a81fa9afa4d32.png",[],{"type":18,"tag":26,"props":137,"children":138},{},[139],{"type":24,"value":140},"本次技术分享通过理论与实践深度结合的形式，为开发者搭建了一条基于昇思MindSpore从模型原理到端侧部署的完整技术路径。开发者不仅掌握了Janus-Pro双编码器架构的创新逻辑，更通过香橙派开发板亲身体验了昇腾软硬件协同优化的全流程。昇思MindSpore联合香橙派，助力开发者创新AI算法，加速落地应用，未来昇思MindSpore将联合昇腾社区与香橙派，提供更丰富完善的模型案例与课程，欢迎感兴趣的开发者一同参与，基于昇思MindSpore+香橙派AIpro构建昇腾软硬件协同的模型和应用。",{"title":7,"searchDepth":142,"depth":142,"links":143},4,[],"markdown","content:activities:zh:3643.md","content","activities/zh/3643.md","activities/zh/3643","md",1776506028508]