[{"data":1,"prerenderedAt":565},["ShallowReactive",2],{"content-query-EGvKqyw3bs":3},{"_path":4,"_dir":5,"_draft":6,"_partial":6,"_locale":7,"title":8,"description":9,"date":10,"cover":11,"type":12,"body":13,"_type":559,"_id":560,"_source":561,"_file":562,"_stem":563,"_extension":564},"/news/zh/3792","zh",false,"","社区月报 | Nature子刊重磅发布！昇思使能首个亿级细胞大模型诞生，0day支持主流模型","昇思MindSpore2025年6月份总结","2025-07-10","https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/07/11/4557638b89a84b5ab64d9623a910a438.png","news",{"type":14,"children":15,"toc":556},"root",[16,24,37,45,50,55,60,65,73,78,83,102,110,120,125,133,143,148,153,164,169,178,187,194,202,211,216,221,230,237,246,251,256,261,268,273,282,292,297,302,309,319,324,329,354,359,366,376,381,388,396,405,410,415,422,430,439,444,449,456,465,470,477,486,491,496,505,512,521,526,533,546,551],{"type":17,"tag":18,"props":19,"children":21},"element","h1",{"id":20},"社区月报-nature子刊重磅发布昇思使能首个亿级细胞大模型诞生0day支持主流模型",[22],{"type":23,"value":8},"text",{"type":17,"tag":25,"props":26,"children":27},"p",{},[28],{"type":17,"tag":29,"props":30,"children":31},"strong",{},[32],{"type":17,"tag":33,"props":34,"children":35},"em",{},[36],{"type":23,"value":9},{"type":17,"tag":25,"props":38,"children":39},{},[40],{"type":17,"tag":29,"props":41,"children":42},{},[43],{"type":23,"value":44},"一、概述",{"type":17,"tag":25,"props":46,"children":47},{},[48],{"type":23,"value":49},"2025年6月，昇思MindSpore开源社区持续保持着高速发展的态度，在技术创新、社区运作等方面都取得了显著的成果。",{"type":17,"tag":25,"props":51,"children":52},{},[53],{"type":23,"value":54},"基于昇思MindSpore AI框架与昇腾AI硬件训练的CellFM模型——全球首个预训练规模达1亿人类细胞的转录组学基础模型在《Nature Communications》发表。其次，昇思MindSpore不断提升HuggingFace Transformers类模型接入vllm-mindspore框架效率，基于vLLM推理框架与MindSpore One套件，0day支持MiniCPM4.0双尺寸模型快速适配。",{"type":17,"tag":25,"props":56,"children":57},{},[58],{"type":23,"value":59},"在社区活动方面，昇思MindSpore社区通过学术基金、call for demo第二期等活动推动AI技术在不同场景中的深度应用，助力开发者成长。在开发者培养方面，昇思MindSpore布道师招募启动，助力开发者实现技术影响力与个人价值的双重提升。高校合作方面，昇思MindSpore联合电子科大开展开源实训营，推动AI技术在教学中的创新应用。",{"type":17,"tag":25,"props":61,"children":62},{},[63],{"type":23,"value":64},"以下是6月昇思MindSpore社区进展的详细报告。",{"type":17,"tag":25,"props":66,"children":67},{},[68],{"type":17,"tag":29,"props":69,"children":70},{},[71],{"type":23,"value":72},"二、社区规模",{"type":17,"tag":25,"props":74,"children":75},{},[76],{"type":23,"value":77},"截至6月底，昇思MindSpore面向全球开发者，凝聚产业力量，持续打造人工智能创新之源。社区秉持开放透明原则，稳步建设开源生态，规模和活跃度得到持续提升。",{"type":17,"tag":25,"props":79,"children":80},{},[81],{"type":23,"value":82},"社区发展了**4.7万+**核心贡献者；",{"type":17,"tag":25,"props":84,"children":85},{},[86,88,93,95,100],{"type":23,"value":87},"累计产生ISSUE共",{"type":17,"tag":29,"props":89,"children":90},{},[91],{"type":23,"value":92},"42.2****K",{"type":23,"value":94},"、PR共",{"type":17,"tag":29,"props":96,"children":97},{},[98],{"type":23,"value":99},"112.7K",{"type":23,"value":101},"。",{"type":17,"tag":25,"props":103,"children":104},{},[105],{"type":17,"tag":29,"props":106,"children":107},{},[108],{"type":23,"value":109},"三、社区大事件",{"type":17,"tag":25,"props":111,"children":112},{},[113,115],{"type":23,"value":114},"**1、**",{"type":17,"tag":29,"props":116,"children":117},{},[118],{"type":23,"value":119},"Nature子刊重磅发布！昇思MindSpore使能全球首个亿级细胞大模型诞生",{"type":17,"tag":25,"props":121,"children":122},{},[123],{"type":23,"value":124},"中山大学、重庆大学等科研院所近日在《Nature Communications》发表开创性研究，正式推出基于昇思MindSpore AI框架与昇腾AI硬件训练的CellFM模型——全球首个预训练规模达1亿人类细胞的转录组学基础模型。这标志着AI技术在生命科学领域取得革命性突破。",{"type":17,"tag":25,"props":126,"children":127},{},[128],{"type":17,"tag":129,"props":130,"children":132},"img",{"alt":7,"src":131},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/07/11/d6d70d665bdb4ebc89a799250cefaeaa.png",[],{"type":17,"tag":25,"props":134,"children":135},{},[136,138],{"type":23,"value":137},"**2、**",{"type":17,"tag":29,"props":139,"children":140},{},[141],{"type":23,"value":142},"0day同步！昇思MindSpore框架成功适配面壁MiniCPM4.0模型",{"type":17,"tag":25,"props":144,"children":145},{},[146],{"type":23,"value":147},"6 月 6 日，面壁智能正式发布并开源了端侧「小钢炮」系列最新力作——MiniCPM 4.0 模型，实现了 220 倍极限加速！基于昇思MindSpore框架深度优化，不断提升HuggingFace Transformers类模型接入vllm-mindspore框架效率，同时基于vLLM推理框架与MindSpore One套件协同，实现0day支持MiniCPM4.0双尺寸模型快速适配。欢迎广大开发者下载体验！",{"type":17,"tag":25,"props":149,"children":150},{},[151],{"type":23,"value":152},"MindSpore-vLLM代码仓：",{"type":17,"tag":25,"props":154,"children":155},{},[156],{"type":17,"tag":157,"props":158,"children":162},"a",{"href":159,"rel":160},"https://gitee.com/mindspore/vllm-mindspore",[161],"nofollow",[163],{"type":23,"value":159},{"type":17,"tag":25,"props":165,"children":166},{},[167],{"type":23,"value":168},"体验指导链接：",{"type":17,"tag":25,"props":170,"children":171},{},[172],{"type":17,"tag":157,"props":173,"children":176},{"href":174,"rel":175},"https://modelers.cn/models/MindSpore-Lab/MiniCPM4-8B",[161],[177],{"type":23,"value":174},{"type":17,"tag":25,"props":179,"children":180},{},[181],{"type":17,"tag":157,"props":182,"children":185},{"href":183,"rel":184},"https://modelers.cn/models/MindSpore-Lab/MiniCPM4-0.5B",[161],[186],{"type":23,"value":183},{"type":17,"tag":25,"props":188,"children":189},{},[190],{"type":17,"tag":129,"props":191,"children":193},{"alt":7,"src":192},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/07/11/14eab6f13429465587d1727805cd5c2a.png",[],{"type":17,"tag":25,"props":195,"children":196},{},[197],{"type":17,"tag":29,"props":198,"children":199},{},[200],{"type":23,"value":201},"四、社区动态",{"type":17,"tag":25,"props":203,"children":204},{},[205,206],{"type":23,"value":114},{"type":17,"tag":29,"props":207,"children":208},{},[209],{"type":23,"value":210},"课程发布 | 玩转昇思 x 昇腾开发板， 轻松掌握DeepSeek大模型训推全流程！",{"type":17,"tag":25,"props":212,"children":213},{},[214],{"type":23,"value":215},"昇思MindSpore技术公开课系列课程上新！《昇思+昇腾开发板：软硬结合玩转DeepSeek开发实战》课程，指导开发者基于昇思在昇腾开发板上玩转大模型开发、微调、推理、性能调优全流程。",{"type":17,"tag":25,"props":217,"children":218},{},[219],{"type":23,"value":220},"课程学习链接：",{"type":17,"tag":25,"props":222,"children":223},{},[224],{"type":17,"tag":157,"props":225,"children":228},{"href":226,"rel":227},"https://www.hiascend.com/developer/courses/detail/1925362775376744449",[161],[229],{"type":23,"value":226},{"type":17,"tag":25,"props":231,"children":232},{},[233],{"type":17,"tag":129,"props":234,"children":236},{"alt":7,"src":235},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/07/11/1cacfdb4ce694567ba9febde9fb60d9b.png",[],{"type":17,"tag":25,"props":238,"children":239},{},[240,241],{"type":23,"value":137},{"type":17,"tag":29,"props":242,"children":243},{},[244],{"type":23,"value":245},"昇思布道师招募！联合华为开发者布道师，双认证加持！",{"type":17,"tag":25,"props":247,"children":248},{},[249],{"type":23,"value":250},"近期，昇思MindSpore布道师计划 联合 华为开发者布道师计划 助力布道师AI布道启航！即日起，成功申请成为 昇思MindSpore布道师 即可同步获得 华为开发者布道师 双认证资质，开启技术布道与生态共建的全新篇章！",{"type":17,"tag":25,"props":252,"children":253},{},[254],{"type":23,"value":255},"无论你是高校师生、企业工程师，还是开源社区贡献者，只要你怀揣技术信仰，渴望用AI改变世界，这里就是你的舞台！",{"type":17,"tag":25,"props":257,"children":258},{},[259],{"type":23,"value":260},"扫描下方二维码了解详情：",{"type":17,"tag":25,"props":262,"children":263},{},[264],{"type":17,"tag":129,"props":265,"children":267},{"alt":7,"src":266},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/07/11/bb7727f9fb1c4e80a816be9f5ba0be80.png",[],{"type":17,"tag":25,"props":269,"children":270},{},[271],{"type":23,"value":272},"原文链接：",{"type":17,"tag":25,"props":274,"children":275},{},[276],{"type":17,"tag":157,"props":277,"children":280},{"href":278,"rel":279},"https://mp.weixin.qq.com/s/4TSu9TPzDznbpgJ07-gztg",[161],[281],{"type":23,"value":278},{"type":17,"tag":25,"props":283,"children":284},{},[285,287],{"type":23,"value":286},"**3、**",{"type":17,"tag":29,"props":288,"children":289},{},[290],{"type":23,"value":291},"昇思MindSpore SIG中心正式上线",{"type":17,"tag":25,"props":293,"children":294},{},[295],{"type":23,"value":296},"6月，昇思社区SIG中心(mindspore.cn/sig) 全新上线,进入【SIG会议日历】，即可轻松获得昇思社区SIG（特别兴趣小组）的最新动态!",{"type":17,"tag":25,"props":298,"children":299},{},[300],{"type":23,"value":301},"开发者可自由选择感兴趣的会议，一键报名，零距离交流社区工作，高效参与社区贡献。",{"type":17,"tag":25,"props":303,"children":304},{},[305],{"type":17,"tag":129,"props":306,"children":308},{"alt":7,"src":307},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/07/11/5bd2e77514d84cae8429ec9ae2c7cae1.png",[],{"type":17,"tag":25,"props":310,"children":311},{},[312,314],{"type":23,"value":313},"**4、**",{"type":17,"tag":29,"props":315,"children":316},{},[317],{"type":23,"value":318},"加速科研创新，GAITC2025展示多项昇思MindSpore学术基金优秀成果",{"type":17,"tag":25,"props":320,"children":321},{},[322],{"type":23,"value":323},"2025年6月6日至8日，全球人工智能技术大会(GAITC2025)在杭州未来科技城盛大召开。大会开幕式上，中国工程院院士、中国人工智能学会理事长戴琼海和国际欧亚科学院院士田奇共同为CAAI-昇思MindSpore/昇腾CANN学术基金优秀项目颁奖。",{"type":17,"tag":25,"props":325,"children":326},{},[327],{"type":23,"value":328},"本次获奖项目涵盖人工智能多个前沿领域，包括：",{"type":17,"tag":330,"props":331,"children":332},"ul",{},[333,339,344,349],{"type":17,"tag":334,"props":335,"children":336},"li",{},[337],{"type":23,"value":338},"中山大学陈武辉教授团队《LongUMoE：基于统一内存的MoE长序列推理框架》",{"type":17,"tag":334,"props":340,"children":341},{},[342],{"type":23,"value":343},"中山大学李冠彬教授团队《基于参数化隐式表达的数字说话人视频生成》",{"type":17,"tag":334,"props":345,"children":346},{},[347],{"type":23,"value":348},"北京安德医智刘盼团队《基于复合策略的多模态VIT模型推理加速方案》",{"type":17,"tag":334,"props":350,"children":351},{},[352],{"type":23,"value":353},"北京航空航天大学孙庆赟团队《基于MindSpore的PDE基础模型构建研究》",{"type":17,"tag":25,"props":355,"children":356},{},[357],{"type":23,"value":358},"这些创新成果充分展现昇思MindSpore生态在推动AI技术创新方面的积极作用，为人工智能技术发展提供了重要支撑。",{"type":17,"tag":25,"props":360,"children":361},{},[362],{"type":17,"tag":129,"props":363,"children":365},{"alt":7,"src":364},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/07/11/6f13087db57244b7a8319ea8c9987ccb.png",[],{"type":17,"tag":25,"props":367,"children":368},{},[369,371],{"type":23,"value":370},"**5、**",{"type":17,"tag":29,"props":372,"children":373},{},[374],{"type":23,"value":375},"电子科大开源实训营 | 昇思MindSpore带你开启LLM应用开发创新之旅",{"type":17,"tag":25,"props":377,"children":378},{},[379],{"type":23,"value":380},"“开源筑梦”为主题的2025年电子科技大学国家大学科技园暑期学校(十三期)将在7月14日-21日举行，昇思MindSpore布道师洪子淇受邀担任本次暑期学校昇思MindSpore实训课的主讲嘉宾。她将以《昇思MindSpore 与大语言模型(LLM)应用实践》为主题，带领开发者完成从理论认知(Transformer 架构、LLM 原理)到动手实践(手写数字识别、Agent+RAG 问答系统搭建)的全流程学习，高效提升 AI应用开发能力。",{"type":17,"tag":25,"props":382,"children":383},{},[384],{"type":17,"tag":129,"props":385,"children":387},{"alt":7,"src":386},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/07/11/62203127ba1f45e6a8925f6e66119a7b.png",[],{"type":17,"tag":25,"props":389,"children":390},{},[391],{"type":17,"tag":29,"props":392,"children":393},{},[394],{"type":23,"value":395},"五、社区案例",{"type":17,"tag":25,"props":397,"children":398},{},[399,400],{"type":23,"value":114},{"type":17,"tag":29,"props":401,"children":402},{},[403],{"type":23,"value":404},"使能油气勘探10倍级降本增效，东方物探联合人大高瓴学院发布基于昇思打造的全波形反演模型",{"type":17,"tag":25,"props":406,"children":407},{},[408],{"type":23,"value":409},"由中石油东方物探、人大高瓴人工智能学院、华为组成的产学研团队，联合开发基于昇思MindSpore AI4S使能套件的FWI新一代解决方案，实现地震波求解效率量级提升，推动油气勘探“AI换芯”进程。",{"type":17,"tag":25,"props":411,"children":412},{},[413],{"type":23,"value":414},"该研究基于MindSpore自动微分框架，构建可微分正演流程，从而在反演中省去传统伴随方法波场反传过程，直接自动求取速度模型梯度修正量，减少了约50%计算量的同时大大简化了计算流程。",{"type":17,"tag":25,"props":416,"children":417},{},[418],{"type":17,"tag":129,"props":419,"children":421},{"alt":7,"src":420},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/07/11/7ece0c7cd860483ab160c35323f205b4.png",[],{"type":17,"tag":25,"props":423,"children":424},{},[425],{"type":17,"tag":29,"props":426,"children":427},{},[428],{"type":23,"value":429},"六、技术分享",{"type":17,"tag":25,"props":431,"children":432},{},[433,434],{"type":23,"value":114},{"type":17,"tag":29,"props":435,"children":436},{},[437],{"type":23,"value":438},"昇思MindSpore论文 | 基于GRPO的图像描述算法",{"type":17,"tag":25,"props":440,"children":441},{},[442],{"type":23,"value":443},"目前图像描述算法的训练采用SCST强化学习方法解决暴露偏差问题（训练与测试输入不一致）。但SCST算法目前存在一定的局限性，其仅依赖单一贪心解码作为基线，可能导致优势估计方差高，并且仅参考贪心解码，生成多样性受限，而缺乏KL散度的约束也容易导致训练崩溃。",{"type":17,"tag":25,"props":445,"children":446},{},[447],{"type":23,"value":448},"昇思MindSpore开发者提出基于GRPO强化学习的图像描述新方法，有效解决传统SCST训练的局限性。该方案利用MindSpore框架，为输入图像生成多个候选描述，计算每个描述的组内优势，结合KL散度，在保证准确性和稳定性的同时不断优化模型。",{"type":17,"tag":25,"props":450,"children":451},{},[452],{"type":17,"tag":129,"props":453,"children":455},{"alt":7,"src":454},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/07/11/2dd693713cd84ca697cd594d47a40ca0.png",[],{"type":17,"tag":25,"props":457,"children":458},{},[459,460],{"type":23,"value":137},{"type":17,"tag":29,"props":461,"children":462},{},[463],{"type":23,"value":464},"昇思MindSpore论文 | 含噪量子本征求解器",{"type":17,"tag":25,"props":466,"children":467},{},[468],{"type":23,"value":469},"针对当前中等规模量子设备受固有噪声影响导致基态能量计算精度不足的问题，昇思MindSpore 开发者们基于矩阵乘积态（Matrix Product States, MPS）提出了新颖的变分量子本征求解算法MPS-VQE。该算法通过优化电路结构设计、改进参数初始化与测量策略，并集成专用误差缓解技术，显著提升了在去极化噪声、热弛豫噪声和比特翻转噪声等复杂环境下的计算精度。基于MindSpore Quantum框架的仿真实验验证，该方案为噪声量子计算环境下的精确模拟提供了有效解决方案。",{"type":17,"tag":25,"props":471,"children":472},{},[473],{"type":17,"tag":129,"props":474,"children":476},{"alt":7,"src":475},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/07/11/1992f4ff752e4ee0b782d790bf498fe0.png",[],{"type":17,"tag":25,"props":478,"children":479},{},[480,481],{"type":23,"value":286},{"type":17,"tag":29,"props":482,"children":483},{},[484],{"type":23,"value":485},"Whisper加速实战：教你用MindSpore Profiler为推理提速",{"type":17,"tag":25,"props":487,"children":488},{},[489],{"type":23,"value":490},"由 OpenAI 开发的多语言语音识别模型Whisper，一经开源受到开发者广泛关注和使用。本文将系统分享我们在MindSpore 2.5.0 + MindSpore NLP 0.4.0环境下，通过引入FlashAttention 2 与优化Conv1D，借助MindSpore Profiler[1]精准定位瓶颈，最终将Whisper模型推理耗时压缩至60秒以内的全过程。",{"type":17,"tag":25,"props":492,"children":493},{},[494],{"type":23,"value":495},"目前该模型已上线模力方舟，点击阅读原文可直接体验。",{"type":17,"tag":25,"props":497,"children":498},{},[499],{"type":17,"tag":157,"props":500,"children":503},{"href":501,"rel":502},"https://ai.gitee.com/serverless-api/packages/1495?model=whisper-large-v3&package=1495",[161],[504],{"type":23,"value":501},{"type":17,"tag":25,"props":506,"children":507},{},[508],{"type":17,"tag":129,"props":509,"children":511},{"alt":7,"src":510},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/07/11/e15ab543325241bda284fdf35b6d57f8.png",[],{"type":17,"tag":25,"props":513,"children":514},{},[515,516],{"type":23,"value":313},{"type":17,"tag":29,"props":517,"children":518},{},[519],{"type":23,"value":520},"昇思MindSpore论文 | 一种用于文本分类的双向上下文感知测试时学习方法",{"type":17,"tag":25,"props":522,"children":523},{},[524],{"type":23,"value":525},"本文提出了一种动态双向上下文感知的测试时学习方法——Dynamic Bidirectional Elman Attention Network（DBEAN），用于在资源受限场景下实现高效、准确的文本分类。该方法依托MindSpore深度学习框架提供的动态图机制与高效的Ascend算子支持，DBEAN能够在推理阶段执行低开销的测试时更新策略，进一步增强模型对输入样本的适应性。",{"type":17,"tag":25,"props":527,"children":528},{},[529],{"type":17,"tag":129,"props":530,"children":532},{"alt":7,"src":531},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/07/11/fd1f5856e6094e53bf9a6ae49798c883.png",[],{"type":17,"tag":25,"props":534,"children":535},{},[536],{"type":17,"tag":29,"props":537,"children":538},{},[539,541],{"type":23,"value":540},"七、",{"type":17,"tag":29,"props":542,"children":543},{},[544],{"type":23,"value":545},"感谢每一位朋友、开发者的支持",{"type":17,"tag":25,"props":547,"children":548},{},[549],{"type":23,"value":550},"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。",{"type":17,"tag":25,"props":552,"children":553},{},[554],{"type":23,"value":555},"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":557,"depth":557,"links":558},4,[],"markdown","content:news:zh:3792.md","content","news/zh/3792.md","news/zh/3792","md",1776506089493]