Are you sure you want to delete this task? Once this task is deleted, it cannot be recovered.
chenzeng8 6d9834a6d7 | 1 year ago | |
---|---|---|
.idea | 1 year ago | |
mxTuningKit | 1 year ago | |
src | 1 year ago | |
.gitignore | 1 year ago | |
LICENSE | 1 year ago | |
README.md | 1 year ago | |
requirements.txt | 1 year ago |
Pangu-Alpha由以鹏城实验室为首的技术团队联合攻关,首次基于“鹏城云脑Ⅱ”和国产MindSpore框架的自动混合并行模式实现在2048卡算力集群上的大规模分布式训练,训练出业界首个2000亿参数以中文为核心的预训练生成语言模型。鹏程·盘古α预训练模型支持丰富的场景应用,在知识问答、知识检索、知识推理、阅读理解等文本生成领域表现突出,具备很强的小样本学习能力。
套件采用云端本地统一架构, 提供一键微调、评估、推理、在线部署 等能力,提供 命令行 和 API接口 两种调用形式,可支持大模型平台的建设与使用。
pip install TuningKit/Ascend_mindxsdk_mxTuningKit-3.0.0-py3-none-any.whl
pip install TuningKit/Ascend_mindxsdk_mxFoundationModel-1.0.1.RC2.b001-py3-none-any.whl
下游任务 | 任务类型 | 论文精度(pangu-2B6) | 复现精度(pangu-2B6) | 样本类型 |
---|---|---|---|---|
AFQMC | 文本相似度 | acc=64.62% | acc=68.9% | one-shot |
CMRC2017 | 中文阅读理解(填空型阅读理解) | acc=38.00% | acc=37.5% | one-shot |
TNEWS | 新闻短文本分类 | acc=57.95% | acc=60.18% | one-shot |
CMRC2018 | 中文阅读理解(句子级填空型阅读理解) | Em/F1=1.21/16.65 | Em/F1=1.06/16.53 | zero-shot |
WebQA | 中文问答 | Em/F1=24/33.94 | Em/F1=23.77/33.86 | few_shot |
IFLYTEK | 应用长文本分类 | acc=74.26% | acc=73.72% | zero-shot |
PD&CFT | 中文阅读理解(填空型阅读理解) | PD: acc=58.05% CFT: acc=42.39% |
PD: acc=58.56% CFT: acc=42.31% |
zero-shot |
DUREADER | 中文阅读理解(知识问答) | Rouge-l=21.07% | Rouge-l=28.98% | zero_shot |
CMRC2019 | 中文阅读理解(片段抽取任务) | acc=67.89% | acc=67.93% | zero-shot |
CSL | 论文关键词识别 | acc=50.9% | acc=52.1% | one-shot |
目前下游任务微调支持8卡执行
评估和推理支持单卡、4卡和8卡执行
No Description
Python Shell Markdown other
Dear OpenI User
Thank you for your continuous support to the Openl Qizhi Community AI Collaboration Platform. In order to protect your usage rights and ensure network security, we updated the Openl Qizhi Community AI Collaboration Platform Usage Agreement in January 2024. The updated agreement specifies that users are prohibited from using intranet penetration tools. After you click "Agree and continue", you can continue to use our services. Thank you for your cooperation and understanding.
For more agreement content, please refer to the《Openl Qizhi Community AI Collaboration Platform Usage Agreement》