Are you sure you want to delete this task? Once this task is deleted, it cannot be recovered.
You can not select more than 25 topics
Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.
sunlongwu
1e6b350aa3
|
4 months ago | |
|---|---|---|
| TopsRider_samples | 4 months ago | |
| 1.sh | 5 months ago | |
| README.md | 5 months ago | |
| run_bert_large.py | 4 months ago | |
| run_bert_large_8card.py | 4 months ago | |
| run_deeplabv3.py | 5 months ago | |
| run_deeplabv3_8card.py | 4 months ago | |
| run_transformer.py | 5 months ago | |
| run_transformer_8card.py | 5 months ago | |
| run_wide_deep_8card.py | 4 months ago | |
README.md
paddle_gcu_2.4.136
本项目以GCU + PaddlePaddle2.5.1框架单卡/多卡 deeplabv3、wide_deep、bert_large、transformer等基础示例，旨在为AI开发者提供启智CV、NLP等多场景的训练参考。
- 派生项目
- 派生项目指创建一个示例代码仓的副本到你的账户下
- 你的派生项目与原示例代码仓为两个独立项目，各自的改动不会互相影响，除非你向原项目提出代码变更申请
gcu_paddlepaddle_modelzoo页面，点击 “派生”，然后按照提示填写内容，选择 “派生项目”。
- 创建训练任务
以单卡bert_large训练为例，选择 “云脑”->“训练任务”->“新建训练任务”。
“智算网络集群” -> “GCU” -> 按需选择“镜像/启动文件/数据集/资源规格” -> “新建训练任务”。
镜像：gcu2.4.136_ubuntu20.04_py3_pytorch_paddle
启动文件：run_bert_large，8卡则填写“ run_bert_large_8card.py”，其他模型以此类推，资源规格根据启动文件选择单卡或8卡。
数据集：paddle_bert_large.zip（选择数据集->公开数据集->搜索“gcu”->选择GCU Dataset->勾选paddle_bert_large.zip），数据集与模型对应关系参考如下表格：
数据集 对应模型
wide_deep criteo.zip
Transformer transformer_paddle.zip
deeplabv3 VOC2012_pytorch_deepLabV3.zip
bert_large paddle_bert_large.zip
No Description
Python Roff Text Gettext Catalog C++ other
