Skip to content

DNN_Partition辅助工具,用于对pytorch模型进行简单的性能分析以及支持模型切分

Notifications You must be signed in to change notification settings

FENP/pytorchtool

Repository files navigation

pytorchtool

DNN_Partition辅助工具,用于对pytorch模型进行简单的性能分析

使用步骤

  1. 对一个初始化后的模型(已加载权重文件)使用save_model函数分别将各层权重保存到**“./model_weight/”**
  2. 对一个未初始化的模型,使用with Profile(model) as p,然后执行模型推理。模型分析被保存在self.information中,使用p.printCsv可以以csv文件格式输出。

添加了示例(example.py classes.py) 已验证

下载模型权重文件

# AlexNet
mkdir alexnet
cd ./alexnet
wget https://download.pytorch.org/models/alexnet-owt-7be5be79.pth
# inceptionV3
mkdir inception_v3
cd ./inception_v3
wget https://download.pytorch.org/models/inception_v3_google-1a9a5a14.pth

性能分析

1. 保存各层权重文件

doPrepare=True,执行python example.py

2. 获取各层性能参数

doPrepare=False , doProf=True,执行python example.py

About

DNN_Partition辅助工具,用于对pytorch模型进行简单的性能分析以及支持模型切分

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages