docker安装GPU版pytorch

1. 安装nvidia驱动

2. 安装nvidia-docker

参考https://blog.csdn.net/u011622208/article/details/103458951

3. 安装GPU版pytorch

1. 安装python，pip，numpy

# 首先更新源
$ apt-get update

# 查看python版本
$ apt-cache madison python
# 选择一个版本，比如 2.7.12-1~16.04

# 安装python 
$ apt-get install python=2.7.12-1~16.04
# 安装pip, 如果是python2.7，最好不要更新pip
$ apt-get install python-pip

# 安装numpy
apt-get install python-numpy

2. 安装pytorch
   pip --default-timeout=100 install torch==0.4.1 -i https://pypi.mirrors.ustc.edu.cn/simple/
   –default-timeout:设置超时的时间
   -i: 设置用国内源

3. 测试

root@aefab7b2cd58:~# python
Python 2.7.12 (default, Oct  8 2019, 14:14:10) 
[GCC 5.4.0 20160609] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> torch.cuda.is_available()
True
>>> 

生成镜像(可选)

docker commit -m 'GPU pytorch' -a 'chenjun' aefab7 chenjun/pytorch041
```

## reference
1.[https://blog.csdn.net/mumoDM/article/details/82503022](https://blog.csdn.net/mumoDM/article/details/82503022)