[{"data":1,"prerenderedAt":102},["ShallowReactive",2],{"content-query-nkfwUO9gYw":3},{"_path":4,"_dir":5,"_draft":6,"_partial":6,"_locale":7,"title":8,"description":9,"date":10,"cover":11,"type":12,"body":13,"_type":96,"_id":97,"_source":98,"_file":99,"_stem":100,"_extension":101},"/news/zh/3690","zh",false,"","持续优化训练性能和易用性！MindSpore MOE专家大模型预训练技术研讨成功举办","MindSpore Developer Day 2025于2025年4月12日在杭州举办。","2025-04-16","https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/04/18/871dd14b3ff04a0fb7e80db26f81c242.png","news",{"type":14,"children":15,"toc":93},"root",[16,24,30,35,43,48,55,60,67,72,79,84],{"type":17,"tag":18,"props":19,"children":21},"element","h1",{"id":20},"持续优化训练性能和易用性mindspore-moe专家大模型预训练技术研讨成功举办",[22],{"type":23,"value":8},"text",{"type":17,"tag":25,"props":26,"children":27},"p",{},[28],{"type":23,"value":29},"MindSpore Developer Day 2025于2025年4月12日在杭州举办。本次大会由昇思MindSpore开源社区主办，聚焦AI框架技术创新与开源开发者生态共建，展示其端到端全流程支持DeepSeek等MoE大模型训练、微调、强化学习与推理部署的技术，以及在科学智能、开发易用性等领域和方向的技术创新与成果。",{"type":17,"tag":25,"props":31,"children":32},{},[33],{"type":23,"value":34},"作为本次大会重要的一环，SIG Gathering吸引了众多开发者的关注。其中，《MindSpore MOE专家大模型预训练技术研讨》专题围绕DeepSeek-V3，开发者们对MindSpore的MOE专家并行、大集群流水负载均衡、性能优化等关键技术支持和演进展开研讨，同时对MindSpore在动态图上的关键能力和进展进行深入的讨论，现场气氛热烈。",{"type":17,"tag":25,"props":36,"children":37},{},[38],{"type":17,"tag":39,"props":40,"children":42},"img",{"alt":7,"src":41},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/04/18/71cecf3bdc8241dd8b29a7f0985494bf.png",[],{"type":17,"tag":25,"props":44,"children":45},{},[46],{"type":23,"value":47},"昇思MindSpore技术专家陈子恒分享了昇思MindSpore对于DeepSeek-V3高性能分布式训练的支持能力，包括多维混合并行的设计和高性能Dropless MoE的实现；同时介绍了昇思MindSpore针对DeepSeek-V3的通信和流水线并行的优化方法。",{"type":17,"tag":25,"props":49,"children":50},{},[51],{"type":17,"tag":39,"props":52,"children":54},{"alt":7,"src":53},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/04/18/7f44e60313c243c7bb3927d0e1493a94.png",[],{"type":17,"tag":25,"props":56,"children":57},{},[58],{"type":23,"value":59},"昇思MindSpore技术专家何青林介绍了MindSpore Transformers大模型套件与使用MindSpore Transformers在大规模集群上拉起DeepSeek-V3预训练的案例，并展示了MindSpore Transformers在预训练方面的功能特性。",{"type":17,"tag":25,"props":61,"children":62},{},[63],{"type":17,"tag":39,"props":64,"children":66},{"alt":7,"src":65},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/04/18/6710549da9bc4206a5351b93269fe42b.png",[],{"type":17,"tag":25,"props":68,"children":69},{},[70],{"type":23,"value":71},"昇思MindSpore技术专家王嘉霖分享了MindSpore动态图兼容Megatron-LM预训练库的方案，介绍了兼容方案的背景和实现设计，并展示了兼容方案实际的使用样例。",{"type":17,"tag":25,"props":73,"children":74},{},[75],{"type":17,"tag":39,"props":76,"children":78},{"alt":7,"src":77},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/04/18/e7feff54478c44a8835a30042c763928.png",[],{"type":17,"tag":25,"props":80,"children":81},{},[82],{"type":23,"value":83},"开发者围绕MindSpore对DeepSeek-V3训练的优化手段和训练的解决方案展开讨论，提出性能优化的方法论、易用性等问题，并明确了持续优化训练性能和调优易用性的工作方向。",{"type":17,"tag":25,"props":85,"children":86},{},[87,91],{"type":17,"tag":39,"props":88,"children":90},{"alt":7,"src":89},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/04/18/0311ad9891814d0f981d2821a6f7be31.png",[],{"type":23,"value":92}," 本次会议上，开发者们围绕昇思MindSpore贡献了大量宝贵想法和建议，昇思MindSpore开源社区的繁荣发展离不开每一位SIG组成员的贡献，从而实现“百花齐放”。",{"title":7,"searchDepth":94,"depth":94,"links":95},4,[],"markdown","content:news:zh:3690.md","content","news/zh/3690.md","news/zh/3690","md",1776506087637]