[{"data":1,"prerenderedAt":352},["ShallowReactive",2],{"content-query-FuBXZH8Wyt":3},{"_path":4,"_dir":5,"_draft":6,"_partial":6,"_locale":7,"title":8,"description":9,"date":10,"cover":11,"type":12,"body":13,"_type":346,"_id":347,"_source":348,"_file":349,"_stem":350,"_extension":351},"\u002Fnews\u002Fzh\u002F2026-6-5","zh",false,"","昇思MindSpore社区 2026 年 5 月份运作报告","2026年5月，昇思MindSpore开源社区持续深耕技术突破与生态共建，在框架能力、社区活动、开发者赋能等方面取得多项重要成果。","2026-5-30","https:\u002F\u002Fobs-mindspore-file.obs.cn-north-4.myhuaweicloud.com\u002Ffile\u002F2025\u002F07\u002F25\u002F199b735845bf4106b44b2035dc97bd39.png","news",{"type":14,"children":15,"toc":326},"root",[16,24,35,42,47,52,57,62,68,73,88,95,100,113,119,126,131,138,143,149,154,161,166,172,177,184,189,195,200,207,212,218,224,229,236,241,247,252,259,264,270,275,282,287,293,298,305,310,316,321],{"type":17,"tag":18,"props":19,"children":21},"element","h1",{"id":20},"昇思mindspore社区-2026-年-5-月份运作报告",[22],{"type":23,"value":8},"text",{"type":17,"tag":25,"props":26,"children":28},"div",{"style":27},"text-align: center;",[29],{"type":17,"tag":30,"props":31,"children":34},"img",{"src":32,"style":33,"alt":7},"\u002Fcategory\u002Finformation\u002Fnews\u002Fbanner\u002F2026-6-5\u002F1.jpg","display: block;margin: 0 auto;max-width:60%",[],{"type":17,"tag":36,"props":37,"children":39},"h2",{"id":38},"一概述",[40],{"type":23,"value":41},"一、概述",{"type":17,"tag":43,"props":44,"children":45},"p",{},[46],{"type":23,"value":9},{"type":17,"tag":43,"props":48,"children":49},{},[50],{"type":23,"value":51},"在技术分享方面，MindSpore Quantum 0.12.0 发布了张量网络模拟器与 QPU 硬件抽象层，让量子算法部署更加便捷。同时，MindSpore Lite 助力语音识别模型实现高效部署，推理时延显著降低。此外，MindSpore HyperParallel 与 LlamaFactory 深度联动，通过通算掩盖、内存优化等大幅提升了 FSDP 训练性能，而MindSpore AKG Agent则提出了基于状态机harness实现的Auto Research工作流，确保长流程的算子迭代优化任务运行稳定。",{"type":17,"tag":43,"props":53,"children":54},{},[55],{"type":23,"value":56},"在社区活动方面，KADC2026大会期间，昇腾AI4S技术论坛聚焦智能体与生命科学应用，昇腾AI框架专题论坛展示了昇思HyperParallel、AKG等组件化能力；湖北工业大学具身机器人AI算法大赛启动，提供开发板与算力券激励；CCF量子挑战赛昇思杯也在持续火热报名中，三场直播与教程助力选手快速上手，昇思正携手开发者共筑开源创新生态。",{"type":17,"tag":43,"props":58,"children":59},{},[60],{"type":23,"value":61},"以下是2026年5月份昇思MindSpore社区进展详细报告。",{"type":17,"tag":36,"props":63,"children":65},{"id":64},"二社区运作",[66],{"type":23,"value":67},"二、社区运作",{"type":17,"tag":43,"props":69,"children":70},{},[71],{"type":23,"value":72},"截至5月底，MindSpore 开源社区稳健运营、生态持续扩容，始终坚守开放协作、技术普惠、共建共治的核心原则，聚焦 AI 框架技术迭代、生态协同与开发者生态建设，社区组织架构高效运转、技术贡献与活跃度双提升，筑牢 AI 开源生态的坚实底座，推动全场景 AI 框架技术与产业应用深度融合。",{"type":17,"tag":74,"props":75,"children":76},"ul",{},[77,83],{"type":17,"tag":78,"props":79,"children":80},"li",{},[81],{"type":23,"value":82},"社区单位会员达 355 家，累计贡献者5.4万+；",{"type":17,"tag":78,"props":84,"children":85},{},[86],{"type":23,"value":87},"社区累计产生ISSUE共91.1K、PR共130.7K。",{"type":17,"tag":25,"props":89,"children":90},{"style":27},[91],{"type":17,"tag":30,"props":92,"children":94},{"src":93,"style":33,"alt":7},"\u002Fcategory\u002Finformation\u002Fnews\u002Fbanner\u002F2026-6-5\u002F2.jpg",[],{"type":17,"tag":43,"props":96,"children":97},{},[98],{"type":23,"value":99},"5月份社区治理体系高效运转，开源发展委员会、技术委员会常态化推进决策与技术统筹，各 SIG 小组协同发力，技术贡献与社区活跃度稳步攀升。",{"type":17,"tag":74,"props":101,"children":102},{},[103,108],{"type":17,"tag":78,"props":104,"children":105},{},[106],{"type":23,"value":107},"5月提交issue 460 个，合入PR 661 个；",{"type":17,"tag":78,"props":109,"children":110},{},[111],{"type":23,"value":112},"新增代码行数超 51.2万行，SIG 专项会议累计召开 7场。",{"type":17,"tag":36,"props":114,"children":116},{"id":115},"三社区活动",[117],{"type":23,"value":118},"三、社区活动",{"type":17,"tag":120,"props":121,"children":123},"h3",{"id":122},"_01-算力驱动科学发现昇腾ai4s助力科研新范式",[124],{"type":23,"value":125},"01 算力驱动科学发现，昇腾AI4S助力科研新范式",{"type":17,"tag":43,"props":127,"children":128},{},[129],{"type":23,"value":130},"5月23日，鲲鹏昇腾开发者大会2026——昇腾AI4S技术论坛在京举行，主题为“算力驱动科学发现”。论坛汇聚企业、高校及科研院所专家，围绕大模型时代下AI4S智能体应用与生命科学等方向展开分享，昇腾展示了在生命科学、材料化学、地球仿真等领域的全面布局，多个原生模型与智能体案例亮相，并为优秀团队颁发了“昇腾AI4S生态贡献优秀奖”。",{"type":17,"tag":25,"props":132,"children":133},{"style":27},[134],{"type":17,"tag":30,"props":135,"children":137},{"src":136,"style":33,"alt":7},"\u002Fcategory\u002Finformation\u002Fnews\u002Fbanner\u002F2026-6-5\u002F3.jpg",[],{"type":17,"tag":43,"props":139,"children":140},{},[141],{"type":23,"value":142},"昇腾将持续夯实软硬件底座，携手各界共建AI4S繁荣生态。未来，昇腾将深化教育、技术与生态协同，助力交叉学科创新，推动AI驱动科学发现的新范式落地。",{"type":17,"tag":120,"props":144,"children":146},{"id":145},"_02-kadc2026-昇腾ai框架专题论坛在京顺利举办",[147],{"type":23,"value":148},"02 KADC2026  |  昇腾AI框架专题论坛在京顺利举办",{"type":17,"tag":43,"props":150,"children":151},{},[152],{"type":23,"value":153},"5月22日，KADC2026“面向超节点，打造开源易用的昇腾AI框架专题论坛”在京顺利举办。会上正式发布 Ascend for PyTorch 社区，并展示了昇思 MindSpore 在分布式并行、图算融合等领域的创新成果，以 HyperParallel、AKG 等组件化能力支撑 PyTorch 等三方生态。LlamaFactory 与 MindSpore HyperParallel 深度联动，一行配置即可开箱加速； AKG 图算融合组件已合入主线，全面赋能大模型图模式训练。",{"type":17,"tag":25,"props":155,"children":156},{"style":27},[157],{"type":17,"tag":30,"props":158,"children":160},{"src":159,"style":33,"alt":7},"\u002Fcategory\u002Finformation\u002Fnews\u002Fbanner\u002F2026-6-5\u002F4.jpg",[],{"type":17,"tag":43,"props":162,"children":163},{},[164],{"type":23,"value":165},"论坛还邀请了多位学术界与产业界专家，分享 API 一致性补齐、多模态大模型及多后端生态标准化建设等前沿实践。面向超节点，昇腾 AI 框架将持续聚焦开源易用，携手开发者共建一致体验与创新技术，让 AI 开发更高效、更便捷。",{"type":17,"tag":120,"props":167,"children":169},{"id":168},"_03-ccf量子计算编程挑战赛昇思杯第八届mindspore-quantum黑客松持续火热报名中",[170],{"type":23,"value":171},"03 CCF量子计算编程挑战赛昇思杯·第八届MindSpore Quantum黑客松持续火热报名中",{"type":17,"tag":43,"props":173,"children":174},{},[175],{"type":23,"value":176},"2026年度CCF量子计算编程挑战赛昇思杯·第八届MindSpore Quantum黑客松持续火热报名中。大赛设置了热身赛、量子组合优化赛道与量子启发算法赛道，并分别于5月9日、5月12日、5月14日举办了三场直播讲解，涵盖赛题解析、编程实践与互动抽奖，帮助选手高效备赛。",{"type":17,"tag":25,"props":178,"children":179},{"style":27},[180],{"type":17,"tag":30,"props":181,"children":183},{"src":182,"style":33,"alt":7},"\u002Fcategory\u002Finformation\u002Fnews\u002Fbanner\u002F2026-6-5\u002F5.jpg",[],{"type":17,"tag":43,"props":185,"children":186},{},[187],{"type":23,"value":188},"官方同步发布了热身赛保姆级通关教程，从环境准备到量子线路搭建、控制门、厄米共轭等基础操作逐题详解，并开放了往届作品、训练营回放及FAQ等学习资源。选手报名后添加小助手即可获取完整指南。欢迎广大量子计算爱好者积极参赛，共赴量子未来。",{"type":17,"tag":120,"props":190,"children":192},{"id":191},"_04-湖北工业大学具身机器人ai算法大赛火热报名中",[193],{"type":23,"value":194},"04 湖北工业大学具身机器人AI算法大赛火热报名中",{"type":17,"tag":43,"props":196,"children":197},{},[198],{"type":23,"value":199},"与此同时，湖北工业大学具身机器人AI算法大赛也正在火热报名中。大赛聚焦具身机器人领域的算法创新，面向广大开发者开放。为激励参赛者，主办方提供了开发板与千元算力券等丰厚奖励，降低硬件门槛，助力选手专注算法突破。",{"type":17,"tag":25,"props":201,"children":202},{"style":27},[203],{"type":17,"tag":30,"props":204,"children":206},{"src":205,"style":33,"alt":7},"\u002Fcategory\u002Finformation\u002Fnews\u002Fbanner\u002F2026-6-5\u002F6.jpg",[],{"type":17,"tag":43,"props":208,"children":209},{},[210],{"type":23,"value":211},"本次比赛是产学研结合的典型案例，欢迎对具身智能、机器人算法感兴趣的开发者踊跃参与。点击阅读原文即可了解报名方式与赛题详情，抓住机会赢取激励，共同推动具身机器人技术创新落地。",{"type":17,"tag":36,"props":213,"children":215},{"id":214},"四技术分享",[216],{"type":23,"value":217},"四、技术分享",{"type":17,"tag":120,"props":219,"children":221},{"id":220},"_01-mindspore-quantum-0120-发布张量网络模拟量子启发式算法族与量子硬件抽象层全面升级",[222],{"type":23,"value":223},"01 MindSpore Quantum 0.12.0 发布：张量网络模拟、量子启发式算法族与量子硬件抽象层全面升级",{"type":17,"tag":43,"props":225,"children":226},{},[227],{"type":23,"value":228},"MindSpore Quantum 发布 0.12.0 版本，围绕“扩展边界”带来三大升级。新增的 mqmps 张量网络模拟器基于矩阵乘积态技术，能以极低内存模拟百比特级低纠缠线路，性能远超传统态矢量方法。QAIA 算法族扩展了 TSB、USB、LSB 三种量化模拟分叉变体，为组合优化问题提供更高效的求解工具。",{"type":17,"tag":25,"props":230,"children":231},{"style":27},[232],{"type":17,"tag":30,"props":233,"children":235},{"src":234,"style":33,"alt":7},"\u002Fcategory\u002Finformation\u002Fnews\u002Fbanner\u002F2026-6-5\u002F7.jpg",[],{"type":17,"tag":43,"props":237,"children":238},{},[239],{"type":23,"value":240},"同时推出的 QPU 硬件抽象层统一了模拟器接口风格，开发者仅需实现采样方法即可对接真实量子硬件，如 Quafu 平台，大幅降低从模拟到真机的迁移成本。新版本打通了从理论验证到硬件部署的关键路径，欢迎广大用户下载体验。",{"type":17,"tag":120,"props":242,"children":244},{"id":243},"_02-语音识别模型部署太难昇思-mindspore-lite-轻松搞定",[245],{"type":23,"value":246},"02 语音识别模型部署太难？昇思 MindSpore Lite 轻松搞定！",{"type":17,"tag":43,"props":248,"children":249},{},[250],{"type":23,"value":251},"语音识别模型部署常面临时延高、国产化适配难等挑战。我们基于开源 FireRedASR 模型，展示了如何借助 MindSpore Lite 在昇腾 Atlas 800I A2 单卡上实现高效部署。通过一键模型转换、融合算子、AOE 自动调优以及 H2D 免拷贝等优化手段，Encoder 与 Decoder 模块的推理时延较开箱性能最高降低 55%，且识别精度与原始模型持平。",{"type":17,"tag":25,"props":253,"children":254},{"style":27},[255],{"type":17,"tag":30,"props":256,"children":258},{"src":257,"style":33,"alt":7},"\u002Fcategory\u002Finformation\u002Fnews\u002Fbanner\u002F2026-6-5\u002F8.jpg",[],{"type":17,"tag":43,"props":260,"children":261},{},[262],{"type":23,"value":263},"整个迁移过程工作量小，集成门槛低，充分验证了 MindSpore Lite 在语音场景的极致性能与易用性。欢迎广大开发者体验并参与社区共建。",{"type":17,"tag":120,"props":265,"children":267},{"id":266},"_03-mindspore-hyperparallel-携手-llamafactory-解锁-fsdp-性能与易用之钥",[268],{"type":23,"value":269},"03 MindSpore HyperParallel 携手 LlamaFactory 解锁 FSDP 性能与易用之钥",{"type":17,"tag":43,"props":271,"children":272},{},[273],{"type":23,"value":274},"大模型训练面临显存与效率的双重挑战。MindSpore HyperParallel 携手 LlamaFactory 深度联动，为 FSDP 后端注入高性能与易用性。通过通信融合、零拷贝、参数预取及梯度规约掩盖等核心技术，在昇腾 NPU 上实现千亿参数训练性能较 Megatron 系列提升约 10%，显存占用节省 5%~10%。",{"type":17,"tag":25,"props":276,"children":277},{"style":27},[278],{"type":17,"tag":30,"props":279,"children":281},{"src":280,"style":33,"alt":7},"\u002Fcategory\u002Finformation\u002Fnews\u002Fbanner\u002F2026-6-5\u002F9.jpg",[],{"type":17,"tag":43,"props":283,"children":284},{},[285],{"type":23,"value":286},"更难得的是，用户只需在原有工作流中添加一行配置即可零侵入启用，同一套脚本还能跨硬件平稳运行。这一联合方案已登陆 LlamaFactory，欢迎开发者体验极致高效的分布式训练新范式。",{"type":17,"tag":120,"props":288,"children":290},{"id":289},"_04-在-mindspore-akg-agent-中的-harness-设计面向长程任务的状态机方法",[291],{"type":23,"value":292},"04 在 MindSpore AKG Agent 中的 Harness 设计：面向长程任务的状态机方法",{"type":17,"tag":43,"props":294,"children":295},{},[296],{"type":23,"value":297},"当 MindSpore AKG Agent 需连续运行数十小时且无人监督时，仅靠提示词约束极易偏离预期。我们提出基于状态机的 harness 设计，以代码逻辑为长程任务划定边界，同时保留 Agent 的自主推理与代码修改能力。",{"type":17,"tag":25,"props":299,"children":300},{"style":27},[301],{"type":17,"tag":30,"props":302,"children":304},{"src":303,"style":33,"alt":7},"\u002Fcategory\u002Finformation\u002Fnews\u002Fbanner\u002F2026-6-5\u002F10.jpg",[],{"type":17,"tag":43,"props":306,"children":307},{},[308],{"type":23,"value":309},"通过主循环状态机与 PLAN 状态机的分工协作，结合落盘 PLAN 文件、Subagent 上下文隔离及端到端评测托管等机制，有效防止跳过评测、遗忘历史等行为，让任务跑得稳、可追溯、可复现。这一范式已在算子性能调优任务中验证，欢迎开发者探索与实践。",{"type":17,"tag":36,"props":311,"children":313},{"id":312},"五感谢每一位朋友开发者的支持",[314],{"type":23,"value":315},"五、感谢每一位朋友、开发者的支持",{"type":17,"tag":43,"props":317,"children":318},{},[319],{"type":23,"value":320},"在此感谢社区伙伴们、可爱的小孢子们以及昇思MindSpore SIG组成员们，因为大家的共同努力及辛勤奉献，昇思MindSpore才能不断成长与发展！同时我们对可能出现的不完善之处向您表示诚挚的歉意，并衷心感谢您的理解与支持。图片图片",{"type":17,"tag":43,"props":322,"children":323},{},[324],{"type":23,"value":325},"未来，昇思MindSpore AI框架将持续致力于打造人工智能创新之源，凝聚产业力量，扎根AI根技术，使能大模型与科学智能，成为AI创新的首选框架。",{"title":7,"searchDepth":327,"depth":327,"links":328},4,[329,331,332,339,345],{"id":38,"depth":330,"text":41},2,{"id":64,"depth":330,"text":67},{"id":115,"depth":330,"text":118,"children":333},[334,336,337,338],{"id":122,"depth":335,"text":125},3,{"id":145,"depth":335,"text":148},{"id":168,"depth":335,"text":171},{"id":191,"depth":335,"text":194},{"id":214,"depth":330,"text":217,"children":340},[341,342,343,344],{"id":220,"depth":335,"text":223},{"id":243,"depth":335,"text":246},{"id":266,"depth":335,"text":269},{"id":289,"depth":335,"text":292},{"id":312,"depth":330,"text":315},"markdown","content:news:zh:2026-6-5.md","content","news\u002Fzh\u002F2026-6-5.md","news\u002Fzh\u002F2026-6-5","md",1781776862596]