[{"data":1,"prerenderedAt":180},["ShallowReactive",2],{"content-query-YBO2yYtCTT":3},{"_path":4,"_dir":5,"_draft":6,"_partial":6,"_locale":7,"title":8,"description":9,"date":10,"cover":11,"type":12,"body":13,"_type":174,"_id":175,"_source":176,"_file":177,"_stem":178,"_extension":179},"/news/zh/3426","zh",false,"","携手昇思，天津大学团队打造生物医学领域大模型“海河·岐伯”","2024年9月19日至21日，华为全联接大会（HC2024）在上海隆重举办。来自天津大学的王鑫老师在大会期间获得“昇腾科研创新卓越贡献者奖”。","2024-10-12","https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2024/10/18/a7cfc9e94dbd413c91a74d76e986852e.png","news",{"type":14,"children":15,"toc":171},"root",[16,24,30,39,44,52,57,69,74,83,88,93,100,104,113,118,127,135,140,147,152,159,166],{"type":17,"tag":18,"props":19,"children":21},"element","h1",{"id":20},"携手昇思天津大学团队打造生物医学领域大模型海河岐伯",[22],{"type":23,"value":8},"text",{"type":17,"tag":25,"props":26,"children":27},"p",{},[28],{"type":23,"value":29},"2024年9月19日至21日，华为全联接大会（HC2024）在上海隆重举办。来自天津大学的王鑫老师在大会期间获得“昇腾科研创新卓越贡献者奖”。作为天津大学智能与计算学部教授、博导、人工智能学院副院长，王鑫老师在古文预训练模型研究积累丰厚，自2022年开始带领团队基于MindSpore原生研发面向古汉语的预训练语言大模型，孵化了生物医学领域大模型—“海河·岐伯”。",{"type":17,"tag":25,"props":31,"children":32},{},[33],{"type":17,"tag":34,"props":35,"children":36},"strong",{},[37],{"type":23,"value":38},"生物医学领域大模型“海河·岐伯”",{"type":17,"tag":25,"props":40,"children":41},{},[42],{"type":23,"value":43},"在“海河·岐伯”前期研究中，针对古文的语法语义结构特性，提出了预训练模型RAC-BERT模型，其在继承Transformer编码器结构的同时，根据古文任务需求，对模型做出相应改进，设计了新的基于部首的预训练任务，调整模型参数，优化网络结构，在大规模古籍语料基础上进行训练，在CCLUE测评基准下相比于BETR-base模型提升了5%以上。该研究成果在知名国际学术会议发表论文。",{"type":17,"tag":25,"props":45,"children":46},{},[47],{"type":17,"tag":48,"props":49,"children":51},"img",{"alt":7,"src":50},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2024/10/18/83ec0a7d87a440729a0c1a1af2c52c59.png",[],{"type":17,"tag":25,"props":53,"children":54},{},[55],{"type":23,"value":56},"论文链接：",{"type":17,"tag":25,"props":58,"children":59},{},[60],{"type":17,"tag":61,"props":62,"children":66},"a",{"href":63,"rel":64},"https://link.springer.com/chapter/10.1007/978-3-031-44696-2%5C_59",[65],"nofollow",[67],{"type":23,"value":68},"https://link.springer.com/chapter/10.1007/978-3-031-44696-2\\_59",{"type":17,"tag":25,"props":70,"children":71},{},[72],{"type":23,"value":73},"昇思MindSpore版开源代码链接：",{"type":17,"tag":25,"props":75,"children":76},{},[77],{"type":17,"tag":61,"props":78,"children":81},{"href":79,"rel":80},"https://github.com/CubeHan/RAC-BERT",[65],[82],{"type":23,"value":79},{"type":17,"tag":25,"props":84,"children":85},{},[86],{"type":23,"value":87},"基于前期的研究成果，为了进一步应用大模型到生物医学领域（中医药领域），对中文基座模型进行微调使其获取识别中医药实体的能力；通过构造中医药知识图谱，处理部分中医药相关知识，设计实现了大模型结合知识图谱的问答系统，该研究成果在国内核心期刊发表论文。",{"type":17,"tag":25,"props":89,"children":90},{},[91],{"type":23,"value":92},"最后，为了进一步提升大语言模型在中医药领域性能，进行中医药领域的二次预训练与指令微调得到“海河·岐伯”大模型。“海河·岐伯”大模型在中医药相关的13个科目选择题共计2237道题的测试中达到最优，7位中医药专家在中医药方面进行的主观评估，“海河·岐伯”达到最优。",{"type":17,"tag":25,"props":94,"children":95},{},[96],{"type":17,"tag":48,"props":97,"children":99},{"alt":7,"src":98},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2024/10/18/b3dfe13b2f924b5ebf52d8a19fa8699d.png",[],{"type":17,"tag":25,"props":101,"children":102},{},[103],{"type":23,"value":56},{"type":17,"tag":25,"props":105,"children":106},{},[107],{"type":17,"tag":61,"props":108,"children":111},{"href":109,"rel":110},"https://link.cnki.net/urlid/11.5602.TP.20230920.1152.004",[65],[112],{"type":23,"value":109},{"type":17,"tag":25,"props":114,"children":115},{},[116],{"type":23,"value":117},"MindSpore版开源代码链接：",{"type":17,"tag":25,"props":119,"children":120},{},[121],{"type":17,"tag":61,"props":122,"children":125},{"href":123,"rel":124},"https://github.com/zhangheyi-1/llmkgqas-tcm",[65],[126],{"type":23,"value":123},{"type":17,"tag":25,"props":128,"children":129},{},[130],{"type":17,"tag":34,"props":131,"children":132},{},[133],{"type":23,"value":134},"昇思AI框架使能“海河·岐伯”训练提速",{"type":17,"tag":25,"props":136,"children":137},{},[138],{"type":23,"value":139},"“海河·岐伯”在训练过程中，充分利用了昇思MindSpore框架的多种并行策略，通过流水并行、数据并行等混合并行训练模式，使得“海河岐伯”的训练效率提升了20%。同时，利用昇思MindSpore提供的混合精度的计算方式，在保持计算精度的同时，提高了计算效率和减少了内存占用。此外，针对开发过程遇到的训练调试问题，充分利用昇思MindSpore Insight可视化调试调优工具，缩短模型精度问题定位的时间。最后，在最新的训练中，采用昇思MindSpore的新特性多级编译技术，开启O1编译选项增加自动算子融合优化，以提升内存复用率，相比于动态图获取更高的性能收益。",{"type":17,"tag":25,"props":141,"children":142},{},[143],{"type":17,"tag":48,"props":144,"children":146},{"alt":7,"src":145},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2024/10/18/bcdf17b527ab4d418a900277daa7910d.png",[],{"type":17,"tag":25,"props":148,"children":149},{},[150],{"type":23,"value":151},"“海河·岐伯”获得中国人工智能学会-昇思MindSpore学术基金支持，并携手天津市人工智能计算中心联合开展产业化探索，通过天大智图公司进行商用。",{"type":17,"tag":25,"props":153,"children":154},{},[155],{"type":17,"tag":48,"props":156,"children":158},{"alt":7,"src":157},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2024/10/18/4512a9cf27694e17a25ef9cca8b05190.png",[],{"type":17,"tag":25,"props":160,"children":161},{},[162],{"type":17,"tag":48,"props":163,"children":165},{"alt":7,"src":164},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2024/10/18/0a4bca45e3e342b3b777b64331bb60d5.png",[],{"type":17,"tag":25,"props":167,"children":168},{},[169],{"type":23,"value":170},"未来，王鑫老师团队将持续携手昇思，构建中医药领域知识图谱，整合古今中医药文献、方剂、药物、疾病等知识，并持续收集整理中医药领域相关数据，对“海河·岐伯”进行迭代训练与优化，提高模型的训练效率和精度。此外，结合中医药“望闻问切”的特性，研发中医药多模态大模型。最终，围绕模型打造落地应用，开发智能化中医诊疗系统，从而实现学术创新到产业落地的闭环，助力学术界与产业界生态繁荣。",{"title":7,"searchDepth":172,"depth":172,"links":173},4,[],"markdown","content:news:zh:3426.md","content","news/zh/3426.md","news/zh/3426","md",1776506084199]