[{"data":1,"prerenderedAt":287},["ShallowReactive",2],{"content-query-NHLKEqY7Pz":3},{"_path":4,"_dir":5,"_draft":6,"_partial":6,"_locale":7,"title":8,"description":9,"date":10,"cover":11,"type":12,"body":13,"_type":281,"_id":282,"_source":283,"_file":284,"_stem":285,"_extension":286},"/version-updates/zh/208","zh",false,"","MindSpore 新版本（0.2.0-alpha）解读","【2020年4月30日】MindSpore发布新版本0.2.0-alpha，用5分钟来快速了解下发布的关键特性吧！","2020-04-30","https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2020/09/21/e75a6c3b298d46b8908951c1587ef96c.png","version-updates",{"type":14,"children":15,"toc":278},"root",[16,24,30,35,44,49,64,69,77,88,96,104,109,114,122,129,134,139,147,152,160,165,172,177,185,190,195,200,207,212,217,225,230,241,246,257,268,273],{"type":17,"tag":18,"props":19,"children":21},"element","h1",{"id":20},"mindspore-新版本020-alpha解读",[22],{"type":23,"value":8},"text",{"type":17,"tag":25,"props":26,"children":27},"p",{},[28],{"type":23,"value":29},"@font-face{ font-family:\"Times New Roman\"; } @font-face{ font-family:\"宋体\"; } @font-face{ font-family:\"SimSun\"; } @font-face{ font-family:\"Symbol\"; } @font-face{ font-family:\"Courier New\"; } @font-face{ font-family:\"Wingdings\"; } @font-face{ font-family:\"Arial\"; } @list l0:level1{ mso-level-number-format:bullet; mso-level-suffix:tab; mso-level-text:\"\"; mso-level-tab-stop:36.0000pt; mso-level-number-position:left; margin-left:36.0000pt;text-indent:-18.0000pt;font-family:Symbol;font-size:10.0000pt;} @list l0:level2{ mso-level-number-format:bullet; mso-level-suffix:tab; mso-level-text:\"o\"; mso-level-tab-stop:72.0000pt; mso-level-number-position:left; margin-left:72.0000pt;text-indent:-18.0000pt;font-family:'Courier New';font-size:10.0000pt;} @list l0:level3{ mso-level-number-format:bullet; mso-level-suffix:tab; mso-level-text:\"\"; mso-level-tab-stop:108.0000pt; mso-level-number-position:left; margin-left:108.0000pt;text-indent:-18.0000pt;font-family:Wingdings;font-size:10.0000pt;} @list l0:level4{ mso-level-number-format:bullet; mso-level-suffix:tab; mso-level-text:\"\"; mso-level-tab-stop:144.0000pt; mso-level-number-position:left; margin-left:144.0000pt;text-indent:-18.0000pt;font-family:Wingdings;font-size:10.0000pt;} @list l0:level5{ mso-level-number-format:bullet; mso-level-suffix:tab; mso-level-text:\"\"; mso-level-tab-stop:180.0000pt; mso-level-number-position:left; margin-left:180.0000pt;text-indent:-18.0000pt;font-family:Wingdings;font-size:10.0000pt;} @list l0:level6{ mso-level-number-format:bullet; mso-level-suffix:tab; mso-level-text:\"\"; mso-level-tab-stop:216.0000pt; mso-level-number-position:left; margin-left:216.0000pt;text-indent:-18.0000pt;font-family:Wingdings;font-size:10.0000pt;} @list l0:level7{ mso-level-number-format:bullet; mso-level-suffix:tab; mso-level-text:\"\"; mso-level-tab-stop:252.0000pt; mso-level-number-position:left; margin-left:252.0000pt;text-indent:-18.0000pt;font-family:Wingdings;font-size:10.0000pt;} @list l0:level8{ mso-level-number-format:bullet; mso-level-suffix:tab; mso-level-text:\"\"; mso-level-tab-stop:288.0000pt; mso-level-number-position:left; margin-left:288.0000pt;text-indent:-18.0000pt;font-family:Wingdings;font-size:10.0000pt;} @list l0:level9{ mso-level-number-format:bullet; mso-level-suffix:tab; mso-level-text:\"\"; mso-level-tab-stop:324.0000pt; mso-level-number-position:left; margin-left:324.0000pt;text-indent:-18.0000pt;font-family:Wingdings;font-size:10.0000pt;} p.MsoNormal{ mso-style-name:正文; mso-style-parent:\"\"; margin:0pt; margin-bottom:.0001pt; mso-layout-grid-align:none; text-autospace:none; mso-pagination:none; line-height:150%; font-family:'Times New Roman'; mso-fareast-font-family:SimSun; font-size:10.5000pt; } span.10{ font-family:'Times New Roman'; } span.15{ font-family:'Times New Roman'; color:rgb(0,0,255); text-decoration:underline; text-underline:single; } p.MsoFooter{ mso-style-name:页脚; margin:0pt; margin-bottom:.0001pt; font-family:Arial; mso-fareast-font-family:SimSun; mso-bidi-font-family:'Times New Roman'; font-size:9.0000pt; } p.MsoHeader{ mso-style-name:页眉; margin:0pt; margin-bottom:.0001pt; layout-grid-mode:char; text-align:justify; text-justify:inter-ideograph; font-family:Arial; mso-fareast-font-family:SimSun; mso-bidi-font-family:'Times New Roman'; font-size:9.0000pt; } span.msoIns{ mso-style-type:export-only; mso-style-name:\"\"; text-decoration:underline; text-underline:single; color:blue; } span.msoDel{ mso-style-type:export-only; mso-style-name:\"\"; text-decoration:line-through; color:red; } table.MsoNormalTable{ mso-style-name:普通表格; mso-style-parent:\"\"; mso-style-noshow:yes; mso-tstyle-rowband-size:0; mso-tstyle-colband-size:0; mso-padding-alt:0.0000pt 5.4000pt 0.0000pt 5.4000pt; mso-para-margin:0pt; mso-para-margin-bottom:.0001pt; mso-pagination:widow-orphan; font-family:'Times New Roman'; font-size:10.0000pt; mso-ansi-language:#0400; mso-fareast-language:#0400; mso-bidi-language:#0400; } @page{mso-page-border-surround-header:no; mso-page-border-surround-footer:no;}@page Section0{ } div.Section0{page:Section0;}",{"type":17,"tag":25,"props":31,"children":32},{},[33],{"type":23,"value":34},"在3月底开源的MindSpore，在紧锣密鼓地一个月开发后，4月30日发布了0.2.0-alpha（后简称0.2）版本。有哪些关键更新呢？让我们用5分钟来快速了解下。",{"type":17,"tag":25,"props":36,"children":37},{},[38],{"type":17,"tag":39,"props":40,"children":41},"strong",{},[42],{"type":23,"value":43},"一、支持Windows使用",{"type":17,"tag":25,"props":45,"children":46},{},[47],{"type":23,"value":48},"首先，给window 10系统用户带来了一个重要的更新，MindSpore 0.2版本支持了Windows安装。官方0.2文档已经正式发布，可以在官网中获取完整的安装指导，包括从编译到安装命令端到端的安装方式。 这里直接给出官网的安装命令，网络好，能力强的自信小伙伴可以直接动手：",{"type":17,"tag":25,"props":50,"children":51},{},[52,54,62],{"type":23,"value":53},"pip install ",{"type":17,"tag":55,"props":56,"children":60},"a",{"href":57,"rel":58},"https://ms-release.obs.cn-north-",[59],"nofollow",[61],{"type":23,"value":57},{"type":23,"value":63}," 4.myhuaweicloud.com/0.2.0-alpha/MindSpore/cpu/x64_windows/mindspore-0.2.0-cp37-cp37m-win_amd64.whl",{"type":17,"tag":25,"props":65,"children":66},{},[67],{"type":23,"value":68},"具体使用上和MindSpore的Linux CPU版本没有任何不同，可以支持LeNet网络进行图片分类，支持的算子等也都是一样的，使用Windows的小伙伴们快来体验吧！",{"type":17,"tag":25,"props":70,"children":71},{},[72],{"type":17,"tag":39,"props":73,"children":74},{},[75],{"type":23,"value":76},"二、多个模型Benchmark发布",{"type":17,"tag":25,"props":78,"children":79},{},[80,82],{"type":23,"value":81},"MindSpore 0.2版本发布了benchmark，包含有机器视觉类常用网络ResNet-50 v1.5、自然语言处理常用网络BERT-Large，分别在1*Ascend 910、8*Ascend 910、16*Ascend 910等不同规格的昇腾AI处理器上训练速度和加速比。图片训练在16*Ascend 910场景下每秒处理27090张图片，自然语言处理在8*Ascend 910场景下每秒处理1613个句子，多机并行训练加速比达到0.95左右。具体信息可以移步 benchmark官方地址：",{"type":17,"tag":55,"props":83,"children":86},{"href":84,"rel":85},"https://www.mindspore.cn/docs/zh-CN/0.2.0-alpha/benchmark.html",[59],[87],{"type":23,"value":84},{"type":17,"tag":25,"props":89,"children":90},{},[91],{"type":17,"tag":39,"props":92,"children":93},{},[94],{"type":23,"value":95},"ResNet50",{"type":17,"tag":25,"props":97,"children":98},{},[99],{"type":17,"tag":100,"props":101,"children":103},"img",{"alt":7,"src":102},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2020/05/07/d2133b74af7f45d39b5f5e94c4f28548.png",[],{"type":17,"tag":25,"props":105,"children":106},{},[107],{"type":23,"value":108},"图1 ResNet50 分布式训练性能数据",{"type":17,"tag":25,"props":110,"children":111},{},[112],{"type":23,"value":113},"说明：以上数据基于华为云AI开发平台ModelArts测试获得，是训练过程整体下沉至Ascend 910 AI处理器执行所得的平均性能。业界其他开源框架数据可参考：ResNet-50 v1.5 for TensorFlow。",{"type":17,"tag":25,"props":115,"children":116},{},[117],{"type":17,"tag":39,"props":118,"children":119},{},[120],{"type":23,"value":121},"BERT",{"type":17,"tag":25,"props":123,"children":124},{},[125],{"type":17,"tag":100,"props":126,"children":128},{"alt":7,"src":127},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2020/05/07/2d799906859348f29eae81da40e35806.png",[],{"type":17,"tag":25,"props":130,"children":131},{},[132],{"type":23,"value":133},"图2 BERT 分布式训练性能数据",{"type":17,"tag":25,"props":135,"children":136},{},[137],{"type":23,"value":138},"说明：以上数据基于华为云AI开发平台ModelArts测试获得，其中网络包含24个隐藏层，句长为128个token，字典表包含21128个token。业界其他开源框架数据可参考：BERT For TensorFlow。",{"type":17,"tag":25,"props":140,"children":141},{},[142],{"type":17,"tag":39,"props":143,"children":144},{},[145],{"type":23,"value":146},"三、可视化工具功能强化",{"type":17,"tag":25,"props":148,"children":149},{},[150],{"type":23,"value":151},"MindSpore 0.1版本发布时给大家带来了不小的惊喜，可视化功能已经包含有模型可视化、模型溯源、数据溯源等功能，帮助小伙伴直观观察网络调试过程。",{"type":17,"tag":25,"props":153,"children":154},{},[155],{"type":17,"tag":39,"props":156,"children":157},{},[158],{"type":23,"value":159},"参数直方图",{"type":17,"tag":25,"props":161,"children":162},{},[163],{"type":23,"value":164},"0.2版本推出可视化组件的新特性“参数分布图”，支持可视化权重和梯度分布情况。支持在网络优化过程中，观察权重&梯度随训练过程的变化，可以帮助小伙伴在网络调优时，判断网络的优化进展。你只需要使用HistogramSummary算子记录感兴趣的权重和梯度，即可在MindSpore可视化组件中的训练看板生成数据的直方图，并观察他们的变化。当出现奇怪的权重数据分布时，可以快速直观得判断训练的健康程度。",{"type":17,"tag":25,"props":166,"children":167},{},[168],{"type":17,"tag":100,"props":169,"children":171},{"alt":7,"src":170},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2020/05/07/10027efb78234afebdcf3d165f7c1330.png",[],{"type":17,"tag":25,"props":173,"children":174},{},[175],{"type":23,"value":176},"图3 可视化工具参数直方图展示",{"type":17,"tag":25,"props":178,"children":179},{},[180],{"type":17,"tag":39,"props":181,"children":182},{},[183],{"type":23,"value":184},"溯源自定义和联动",{"type":17,"tag":25,"props":186,"children":187},{},[188],{"type":23,"value":189},"溯源自定义可以帮助用户记录、查询模型迭代调试开发过程使用的模型超参和数据处理增强。大家只需要在模型训练测试过程添加TrainLineage/EvalLineage callback记录溯源信息，即可在MindInsight模型溯源和数据溯源标签页回溯查看。",{"type":17,"tag":25,"props":191,"children":192},{},[193],{"type":23,"value":194}," 支持通过user_defined_info参数记录自定义溯源信息。",{"type":17,"tag":25,"props":196,"children":197},{},[198],{"type":23,"value":199}," 支持模型参数和数据处理参数的筛选和排序。",{"type":17,"tag":25,"props":201,"children":202},{},[203],{"type":17,"tag":100,"props":204,"children":206},{"alt":7,"src":205},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2020/05/07/d5a4cd69ee1d4c2b838d41ec633deb79.png",[],{"type":17,"tag":25,"props":208,"children":209},{},[210],{"type":23,"value":211},"图4 可视化工具模型溯源页面展示",{"type":17,"tag":25,"props":213,"children":214},{},[215],{"type":23,"value":216},"模型溯源界面指定的筛选条件将会在数据溯源同步展示。反之，数据溯源得到的筛选结果也将在模型溯源界面体现，形成溯源信息查询联动。",{"type":17,"tag":25,"props":218,"children":219},{},[220],{"type":17,"tag":39,"props":221,"children":222},{},[223],{"type":23,"value":224},"四、总结",{"type":17,"tag":25,"props":226,"children":227},{},[228],{"type":23,"value":229},"以上就是MindSpore 0.2版本关键特性。当然新版本还有不少特性，大家可以移步MindSpore ReleaseNote里找到完整介绍：",{"type":17,"tag":25,"props":231,"children":232},{},[233,235],{"type":23,"value":234},"ReleaseNote：",{"type":17,"tag":55,"props":236,"children":239},{"href":237,"rel":238},"https://gitee.com/mindspore/mindspore/blob/master/RELEASE.md",[59],[240],{"type":23,"value":237},{"type":17,"tag":25,"props":242,"children":243},{},[244],{"type":23,"value":245},"MindSpore在不断的发展中，未来还会有更多好用的特性发布，这里也非常欢迎大家下载并体验MindSpore，参与到MindSpore开源社区建设中来。",{"type":17,"tag":25,"props":247,"children":248},{},[249,251],{"type":23,"value":250},"Gitee：",{"type":17,"tag":55,"props":252,"children":255},{"href":253,"rel":254},"https://gitee.com/mindspore/mindspore",[59],[256],{"type":23,"value":253},{"type":17,"tag":25,"props":258,"children":259},{},[260,262],{"type":23,"value":261},"GitHub：",{"type":17,"tag":55,"props":263,"children":266},{"href":264,"rel":265},"https://github.com/mindspore-ai/mindspore",[59],[267],{"type":23,"value":264},{"type":17,"tag":25,"props":269,"children":270},{},[271],{"type":23,"value":272},"官方公众号：MindSpore",{"type":17,"tag":25,"props":274,"children":275},{},[276],{"type":23,"value":277},"官方QQ群：871543436",{"title":7,"searchDepth":279,"depth":279,"links":280},4,[],"markdown","content:version-updates:zh:208.md","content","version-updates/zh/208.md","version-updates/zh/208","md",1776506144138]