[{"data":1,"prerenderedAt":196},["ShallowReactive",2],{"content-query-yMyrPFawqh":3},{"_path":4,"_dir":5,"_draft":6,"_partial":6,"_locale":7,"title":8,"description":9,"date":10,"cover":11,"type":12,"body":13,"_type":190,"_id":191,"_source":192,"_file":193,"_stem":194,"_extension":195},"/news/zh/3235","zh",false,"","6小时完成Llama 3.1-8B适配并开源！MindSpore Transformers大模型套件使能开发者创新","北京时间7月24号, Meta 正式发布 Llama 3.1开源大语言模型, 包含8B、70B 及 405B 参数版本。","2024-07-24","https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2024/07/26/0297902eb57342a5877e5d5064bcde8d.png","news",{"type":14,"children":15,"toc":187},"root",[16,24,30,39,47,57,88,93,98,106,117,125,130,137,145,152,159,166,174,182],{"type":17,"tag":18,"props":19,"children":21},"element","h1",{"id":20},"_6小时完成llama-31-8b适配并开源mindspore-transformers大模型套件使能开发者创新",[22],{"type":23,"value":8},"text",{"type":17,"tag":25,"props":26,"children":27},"p",{},[28],{"type":23,"value":29},"北京时间7月24号, Meta 正式发布 Llama 3.1开源大语言模型, 包含8B、70B 及 405B 参数版本。其中的Llama 3.1 405B模型，在多项基准测试中可以媲美GPT4、Claude3.5等目前最强的闭源大模型。此外，8B 与 70B 参数的 Llama 3.1 模型与其他参数量相似的闭源和开源模型相比，也同样具有竞争力。",{"type":17,"tag":25,"props":31,"children":32},{},[33],{"type":17,"tag":34,"props":35,"children":36},"strong",{},[37],{"type":23,"value":38},"01",{"type":17,"tag":25,"props":40,"children":41},{},[42],{"type":17,"tag":34,"props":43,"children":44},{},[45],{"type":23,"value":46},"6小时完成Llama 3.1适配",{"type":17,"tag":25,"props":48,"children":49},{},[50,55],{"type":17,"tag":34,"props":51,"children":52},{},[53],{"type":23,"value":54},"开发者基于MindSpore Transformers大模型套件，6小时完成Llama3.1-8b的微调及推理的迁移适配，并将代码开源至Gitee代码平台",{"type":23,"value":56},"，面向所有开发者开放体验、调用，后续将同步上传至各大模型社区。",{"type":17,"tag":25,"props":58,"children":59},{},[60,62,71,73,79,81,86],{"type":23,"value":61},"MindSpore Transformers Llama3.1代码仓地址：",{"type":17,"tag":63,"props":64,"children":68},"a",{"href":65,"rel":66},"https://gitee.com/mindspore/mindformers/blob/dev/research/llama3%5C_1",[67],"nofollow",[69],{"type":23,"value":70},"https://gitee.com/mindspore/mindformers/blob/dev/research/llama3\\_1",{"type":23,"value":72}," MindSpore代码仓地址： ",{"type":17,"tag":63,"props":74,"children":77},{"href":75,"rel":76},"https://gitee.com/mindspore/mindspore",[67],[78],{"type":23,"value":75},{"type":23,"value":80}," 开发上，基于MindSpore Transformers大模型套件上",{"type":17,"tag":34,"props":82,"children":83},{},[84],{"type":23,"value":85},"开箱即用的Llama3进行微调和推理部署",{"type":23,"value":87},"，即可得到Llama 3.1。",{"type":17,"tag":25,"props":89,"children":90},{},[91],{"type":23,"value":92},"微调上，开发者利用大模型套件的权重转换工具，一行代码实现huggingface权重一键转化，同时通过使用MindFormers提供的Trainer等高阶易用性接口，仅需修改配置文件即可完成微调适配并成功跑通训练，loss曲线表明微调后训练任务稳定运行。",{"type":17,"tag":25,"props":94,"children":95},{},[96],{"type":23,"value":97},"推理部署上，代码拉取、权重下载、权重转换步骤与微调一致，实现权重一键转换，推理结果显示，基于MindSpore Transformers适配的Llama 3.1模型精度与原生模型精度对齐。",{"type":17,"tag":25,"props":99,"children":100},{},[101],{"type":17,"tag":34,"props":102,"children":103},{},[104],{"type":23,"value":105},"MindSpore Transformers llama3.1开箱流程",{"type":17,"tag":25,"props":107,"children":108},{},[109],{"type":17,"tag":34,"props":110,"children":111},{},[112],{"type":17,"tag":34,"props":113,"children":114},{},[115],{"type":23,"value":116},"微调：",{"type":17,"tag":25,"props":118,"children":119},{},[120],{"type":17,"tag":121,"props":122,"children":124},"img",{"alt":7,"src":123},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2024/07/26/bac29a5d64154953a27a39eb501a2a8d.png",[],{"type":17,"tag":25,"props":126,"children":127},{},[128],{"type":23,"value":129},"模型跑通后，Loss曲线显示微调任务稳定运行。",{"type":17,"tag":25,"props":131,"children":132},{},[133],{"type":17,"tag":121,"props":134,"children":136},{"alt":7,"src":135},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2024/07/26/5f2a3f6f73a14a65bc49bd704e882783.png",[],{"type":17,"tag":25,"props":138,"children":139},{},[140],{"type":17,"tag":34,"props":141,"children":142},{},[143],{"type":23,"value":144},"推理：",{"type":17,"tag":25,"props":146,"children":147},{},[148],{"type":17,"tag":121,"props":149,"children":151},{"alt":7,"src":150},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2024/07/26/fbaedaa349824d258fcdf9ed8224f56f.png",[],{"type":17,"tag":25,"props":153,"children":154},{},[155],{"type":17,"tag":121,"props":156,"children":158},{"alt":7,"src":157},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2024/07/26/6cb31c1cb7fd4d5787af2122df5264f9.png",[],{"type":17,"tag":25,"props":160,"children":161},{},[162],{"type":17,"tag":121,"props":163,"children":165},{"alt":7,"src":164},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2024/07/26/aa58e7a0784848a79b5ea4a5e3551da7.png",[],{"type":17,"tag":25,"props":167,"children":168},{},[169],{"type":17,"tag":34,"props":170,"children":171},{},[172],{"type":23,"value":173},"02",{"type":17,"tag":25,"props":175,"children":176},{},[177],{"type":17,"tag":34,"props":178,"children":179},{},[180],{"type":23,"value":181},"关于MindSpore和MindSpore Transformers",{"type":17,"tag":25,"props":183,"children":184},{},[185],{"type":23,"value":186},"昇思MindSpore是华为开源的首个业界全场景AI融合框架，原生支持AI大模型与科学智能。大模型开发套件MindSpore Transformers基于MindSpore内置的并行技术和组件化设计,构建了大模型训练、微调、评估、推理、部署的全流程开发套件，支持业内主流的Transformer类预训练模型和SOTA下游任务应用，可以帮助用户轻松的实现大模型训练和创新研发。",{"title":7,"searchDepth":188,"depth":188,"links":189},4,[],"markdown","content:news:zh:3235.md","content","news/zh/3235.md","news/zh/3235","md",1776506082166]