ubuntu18.04+cuda10.0+anaconda3安装pytorch记录

1. 首先使用root权限创建环境：
   sudo -s
conda create-n 你的虚拟环境名 python=3.7

2. 查询与自己安装的cuda对应的版本，由于直接在线安装较慢，有以下两种较快的方法：
   强烈推荐方法二，亲测好用
   1）方法一：可以先将torch-1.3.0和torchvision-0.4.1的.whl文件下载好，下载地址：
   https://download.pytorch.org/whl/torch_stable.html
   激活刚刚创建的环境，比如这里我的是：
   source activate pytorch
   进入环境后使用pip命令离线安装torch和torchvision：
   a. 进入下载目录，并安装torch：
   pip3 install torch-1.0.0-cp37-cp37m-linux_x86_64.whl

   (pytorch) root@yy:~# cd '/home/yy/下载' 
   (pytorch) root@yy:~/下载# ls
   Anaconda3-2020.02-Linux-x86_64.sh  cudnn-10.0-linux-x64-v7.4.2.24.tgz
   cuda                               torch-1.0.0-cp37-cp37m-linux_x86_64.whl
   cuda_10.0.130_410.48_linux.run
   (pytorch) root@yy:~/下载# pip3 install torch-1.0.0-cp37-cp37m-linux_x86_64.whl
   WARNING: The directory '/home/yy/.cache/pip' or its parent directory is not owned or is not writable by the current user. The cache has been disabled. Check the permissions and owner of that directory. If executing pip with sudo, you may want sudo's -H flag.
   Processing ./torch-1.0.0-cp37-cp37m-linux_x86_64.whl
   Installing collected packages: torch
   Successfully installed torch-1.0.0
   

   注：如果出现如下Requirement already satisfied

   Requirement already satisfied: pillow>=4.1.1 in /home/yy/anaconda3/envs/pytorch/lib/python3.7/site-packages (from torchvision==0.3.0) (7.2.0)
   Requirement already satisfied: six in /home/yy/anaconda3/envs/pytorch/lib/python3.7/site-packages (from torchvision==0.3.0) (1.15.0)
   

   可以使用 pip3 install --target=目标路径 工具包名字 指定安装路径来解决：
   pip3 install --target=/home/yy/anaconda3/envs/pytorch/lib/python3.7/site-packages torchvision-0.3.0-cp37-cp37m-linux_x86_64.whl
   2）方法二：使用清华源下载pytorch，快速离线安装进入
   https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/linux-64/
   Ctrl+F输入pytorch，找到对应的版本并右键复制连接，打开终端使用wget命令下载
   wget https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/linux-64/pytorch-1.0.1-cuda100py37he554f03_0.tar.bz2
   下载结果如下，存放在主目录中

   (pytorch) root@yy:~# wget https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/linux-64/pytorch-1.0.1-cuda100py37he554f03_0.tar.bz2
   --2020-08-17 12:03:59--  https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/linux-64/pytorch-1.0.1-cuda100py37he554f03_0.tar.bz2
   正在解析主机 mirrors.tuna.tsinghua.edu.cn (mirrors.tuna.tsinghua.edu.cn)... 101.6.8.193, 2402:f000:1:408:8100::1
   正在连接 mirrors.tuna.tsinghua.edu.cn (mirrors.tuna.tsinghua.edu.cn)|101.6.8.193|:443... 已连接。
   已发出 HTTP 请求，正在等待回应... 200 OK
   长度： 361340025 (345M) [application/octet-stream]
   正在保存至: “pytorch-1.0.1-cuda100py37he554f03_0.tar.bz2”
   pytorch-1.0.1-cuda1 100%[++++===============>] 344.60M  4.14MB/s    用时 2m 19ss
   
   2020-08-17 12:07:57 (1.89 MB/s) - 已保存 “pytorch-1.0.1-cuda100py37he554f03_0.tar.bz2” [361340025/361340025])
   

   进入主目录执行安装：
   conda install pytorch-1.0.1-cuda100py37he554f03_0.tar.bz2

   (pytorch) root@yy:~# ls
   anaconda3                                    公共的  图片  音乐
   NVIDIA_CUDA-10.0_Samples                     模板    文档  桌面
   pytorch-1.0.1-cuda100py37he554f03_0.tar.bz2  视频    下载
   (pytorch) root@yy:~# conda install pytorch-1.0.1-cuda100py37he554f03_0.tar.bz2
   
   Downloading and Extracting Packages
   ######################################################################## | 100% 
   Preparing transaction: done
   Verifying transaction: done
   Executing transaction: done
   

   最后安装torchvision：conda install torchvision

3. 检查pytorch是否安装成功：

   (pytorch) root@yy:~# python
   >>> import torch
   >>> print(torch.version)
   <module 'torch.version' from '/home/yy/anaconda3/envs/pytorch/lib/python3.7/site-packages/torch/version.py'>
   >>> print(torch.cuda.is_available())
   True
   

   看到返回值为Ture，说明pytorch安装成功。