[{"data":1,"prerenderedAt":298},["ShallowReactive",2],{"content-query-5M3OOrxXUL":3},{"_path":4,"_dir":5,"_draft":6,"_partial":6,"_locale":7,"title":8,"description":9,"date":10,"cover":11,"type":12,"body":13,"_type":292,"_id":293,"_source":294,"_file":295,"_stem":296,"_extension":297},"/news/zh/3744","zh",false,"","MindSpore成立MindSpore PyNative SIG：推动MindSpore动态图模式的研发与优化","为了提升MindSpore动态图模式的性能、灵活性和易用性，以满足深度学习领域对快速开发、调试、执行等的需求，昇思MindSpore社区正式成立动态图模式框架开发SIG（MindSpore PyNative SIG），专注推动MindSpore动态图（即时执行， Eager Execution）模式的研发与优化。","2025-05-30","https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/06/06/73e9c1faf1c644afbcce0821db5edaac.png","news",{"type":14,"children":15,"toc":289},"root",[16,24,29,38,43,48,53,58,63,71,76,81,92,101,110,118,123,128,133,141,146,151,156,161,169,187,192,201,206,215,220,229,234,242,247,254,262,269,277,282],{"type":17,"tag":18,"props":19,"children":21},"element","h1",{"id":20},"mindspore成立mindspore-pynative-sig推动mindspore动态图模式的研发与优化",[22],{"type":23,"value":8},"text",{"type":17,"tag":25,"props":26,"children":27},"p",{},[28],{"type":23,"value":9},{"type":17,"tag":25,"props":30,"children":31},{},[32],{"type":17,"tag":33,"props":34,"children":35},"strong",{},[36],{"type":23,"value":37},"MindSpore动态图整体架构：",{"type":17,"tag":25,"props":39,"children":40},{},[41],{"type":23,"value":42},"为破解动态图开发灵活性与静态图部署性能不可兼得的行业难题，MindSpore以 “动静结合” 为核心思想重构动态图架构：",{"type":17,"tag":25,"props":44,"children":45},{},[46],{"type":23,"value":47},"1、动静结合的前端表达接口，新增mint模块，功能对齐业界主流AI框架。",{"type":17,"tag":25,"props":49,"children":50},{},[51],{"type":23,"value":52},"2、正向执行使用多级流水加速，弥补数据结构和流程引入的额外开销，提升单算子下发性能。",{"type":17,"tag":25,"props":54,"children":55},{},[56],{"type":23,"value":57},"3、动静结合的执行方式，使用AST和字节码入图方式，实现动静混合执行。",{"type":17,"tag":25,"props":59,"children":60},{},[61],{"type":23,"value":62},"4、自动微分机制复用静态图同一套机制，正反向hook对齐业界主流AI框架，兼顾自动微分灵活性。",{"type":17,"tag":25,"props":64,"children":65},{},[66],{"type":17,"tag":67,"props":68,"children":70},"img",{"alt":7,"src":69},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/06/06/e7ddcdd22bc94b7d95b5dca202f9c2b2.png",[],{"type":17,"tag":25,"props":72,"children":73},{},[74],{"type":23,"value":75},"该设计在保留 Python 实时调试优势的同时，显著提升动态图执行效率，实现 “动态研发、静态部署” 的全场景无缝衔接，为AI模型从实验到落地提供统一技术基座。",{"type":17,"tag":25,"props":77,"children":78},{},[79],{"type":23,"value":80},"MindSpore PyNative项目代码仓模块目录：",{"type":17,"tag":25,"props":82,"children":83},{},[84],{"type":17,"tag":85,"props":86,"children":90},"a",{"href":87,"rel":88},"https://gitee.com/mindspore/mindspore/tree/master/mindspore/ccsrc/pynative",[89],"nofollow",[91],{"type":23,"value":87},{"type":17,"tag":25,"props":93,"children":94},{},[95],{"type":17,"tag":85,"props":96,"children":99},{"href":97,"rel":98},"https://gitee.com/mindspore/mindspore/tree/master/mindspore/ccsrc/pyboost",[89],[100],{"type":23,"value":97},{"type":17,"tag":25,"props":102,"children":103},{},[104],{"type":17,"tag":85,"props":105,"children":108},{"href":106,"rel":107},"https://gitee.com/mindspore/mindspore/tree/master/mindspore/ccsrc/runtime/pynative",[89],[109],{"type":23,"value":106},{"type":17,"tag":25,"props":111,"children":112},{},[113],{"type":17,"tag":33,"props":114,"children":115},{},[116],{"type":23,"value":117},"当前SIG重点工作方向：",{"type":17,"tag":25,"props":119,"children":120},{},[121],{"type":23,"value":122},"1、动态图分发机制开发：Tensor核心数据结构，API分发机制等维护和开发。",{"type":17,"tag":25,"props":124,"children":125},{},[126],{"type":23,"value":127},"2、动态图自动微分机制：自动微分机制，自定义反向，hook机制等微分机制的维护和开发。",{"type":17,"tag":25,"props":129,"children":130},{},[131],{"type":23,"value":132},"3、API接入机制：Tensor API和Functional API的生成和接入机制的维护和开发。",{"type":17,"tag":25,"props":134,"children":135},{},[136],{"type":17,"tag":33,"props":137,"children":138},{},[139],{"type":23,"value":140},"SIG目标：",{"type":17,"tag":25,"props":142,"children":143},{},[144],{"type":23,"value":145},"1、性能优化：提升动态图模式下的API下发效率，降低与静态图的性能差距。",{"type":17,"tag":25,"props":147,"children":148},{},[149],{"type":23,"value":150},"2、功能增强：完善动态图对主流模型的支持。",{"type":17,"tag":25,"props":152,"children":153},{},[154],{"type":23,"value":155},"3、开发者体验：简化API设计，提供更友好的错误提示，降低迁移成本。",{"type":17,"tag":25,"props":157,"children":158},{},[159],{"type":23,"value":160},"4、生态扩展：支持更多硬件后端的接入和扩展。",{"type":17,"tag":25,"props":162,"children":163},{},[164],{"type":17,"tag":33,"props":165,"children":166},{},[167],{"type":23,"value":168},"SIG例会：",{"type":17,"tag":25,"props":170,"children":171},{},[172],{"type":17,"tag":33,"props":173,"children":174},{},[175,177,185],{"type":23,"value":176},"双周例会，北",{"type":17,"tag":33,"props":178,"children":179},{},[180],{"type":17,"tag":33,"props":181,"children":182},{},[183],{"type":23,"value":184},"京时间周四",{"type":23,"value":186},"晚19点。",{"type":17,"tag":25,"props":188,"children":189},{},[190],{"type":23,"value":191},"首次例会录屏：",{"type":17,"tag":25,"props":193,"children":194},{},[195],{"type":17,"tag":85,"props":196,"children":199},{"href":197,"rel":198},"https://www.bilibili.com/video/BV1NB7nzfEXq",[89],[200],{"type":23,"value":197},{"type":17,"tag":25,"props":202,"children":203},{},[204],{"type":23,"value":205},"例会纪要：",{"type":17,"tag":25,"props":207,"children":208},{},[209],{"type":17,"tag":85,"props":210,"children":213},{"href":211,"rel":212},"https://etherpad.mindspore.cn/p/sig-MindSpore-PyNative-meetings",[89],[214],{"type":23,"value":211},{"type":17,"tag":25,"props":216,"children":217},{},[218],{"type":23,"value":219},"订阅邮件：",{"type":17,"tag":25,"props":221,"children":222},{},[223],{"type":17,"tag":85,"props":224,"children":227},{"href":225,"rel":226},"https://www.mindspore.cn/sig/MindSpore%20PyNative",[89],[228],{"type":23,"value":225},{"type":17,"tag":25,"props":230,"children":231},{},[232],{"type":23,"value":233},"（欢迎大家订阅SIG邮件，后续例会将通过邮件通知大家）",{"type":17,"tag":25,"props":235,"children":236},{},[237],{"type":17,"tag":33,"props":238,"children":239},{},[240],{"type":23,"value":241},"SIG成员：",{"type":17,"tag":25,"props":243,"children":244},{},[245],{"type":23,"value":246},"Maintainer 列表",{"type":17,"tag":25,"props":248,"children":249},{},[250],{"type":17,"tag":67,"props":251,"children":253},{"alt":7,"src":252},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/06/06/2a415c073e8c45d49d493302ed4764c2.png",[],{"type":17,"tag":25,"props":255,"children":256},{},[257],{"type":17,"tag":33,"props":258,"children":259},{},[260],{"type":23,"value":261},"Committer列****表",{"type":17,"tag":25,"props":263,"children":264},{},[265],{"type":17,"tag":67,"props":266,"children":268},{"alt":7,"src":267},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/06/06/0d300b52eb284d4ead7428756f6dd975.png",[],{"type":17,"tag":25,"props":270,"children":271},{},[272],{"type":17,"tag":33,"props":273,"children":274},{},[275],{"type":23,"value":276},"加入我们：",{"type":17,"tag":25,"props":278,"children":279},{},[280],{"type":23,"value":281},"欢迎感兴趣的朋友们加入到MindSpore PyNative SIG，共同探讨动态图技术，可以添加小助手微信或扫描微信群二维码，入群讨论。欢迎您的围观和加入！",{"type":17,"tag":25,"props":283,"children":284},{},[285],{"type":17,"tag":67,"props":286,"children":288},{"alt":7,"src":287},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/06/06/b328535f6945480387472cc4a196ca67.png",[],{"title":7,"searchDepth":290,"depth":290,"links":291},4,[],"markdown","content:news:zh:3744.md","content","news/zh/3744.md","news/zh/3744","md",1776506088705]