cuda 11情况下如何配置pytorch 10.2

小灰灰 2022-12-15 03:30 11阅读 0赞

**由于目前pytorch只能支持到10.2的版本，但ubuntu最新的系统驱动直接支持了cuda 11.0， 并且cuda tooklit支持的默认下载也是11.0。**

##  ##

**1. 需要先降低cuda tooklit的版本**

cuda-uninstaller in /usr/local/cuda-11.1/bin
    sudo rm -rf /usr/local/cuda-11.1

来自：[Nidia][]  
或

cd /usr/local/cuda-11.0/bin/
    sudo ./cuda-uninstaller
    sudo rm -rf /usr/local/cuda-11.1

来自：[NVIDIA CUDA Toolkit 11.0 安装与卸载(Linux/Ubuntu)][NVIDIA CUDA Toolkit 11.0 _Linux_Ubuntu]

**2. 之后下载安装cuda 10.2:**

下载地址：[  
CUDA Toolkit 10.2 Download  
][CUDA Toolkit 10.2 Download]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80NTg4NTIzMg_size_16_color_FFFFFF_t_70_pic_center]

wget http://developer.download.nvidia.com/compute/cuda/10.2/Prod/local_installers/cuda_10.2.89_440.33.01_linux.runsudo sh cuda_10.2.89_440.33.01_linux.run

**3. 只安装驱动外的其他内容，安装结束后测试是否成功：**

cd /usr/local/cuda-10.2/samples/1_Utilities/deviceQuery
    sudo make
    sudo ./deviceQuery

出现PASS，代表没有问题

**4. 安装cudnn**  
去[官网][Link 1]选择适合自己的版本，有两种可选：  
![在这里插入图片描述][20201015104330286.png_pic_center]和  
![在这里插入图片描述][20201015104347578.png_pic_center]  
此处我选择的第二个，下载这三个文件：

![在这里插入图片描述][202010151044217.png_pic_center]  
之后使用命令安装：

sudo dpkg -i libcudnn7_7.6.5.32-1+cuda10.2_amd64.deb
    sudo dpkg -i libcudnn7-dev_7.6.5.32-1+cuda10.2_amd64.deb
    sudo dpkg -i libcudnn7-doc_7.6.5.32-1+cuda10.2_amd64.deb

**5. 安装pytorch**  
[官网][Link 2]选择即可，如果网速慢可参考[另一篇文档][Link 3]：  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80NTg4NTIzMg_size_16_color_FFFFFF_t_70_pic_center 1]  
网速比较快，出现下面的内容说明配置正确：  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80NTg4NTIzMg_size_16_color_FFFFFF_t_70_pic_center 2]

**6. 测试**  
输入命令：

python
    >>> import torch
    >>> torch.cuda.is_available()

![在这里插入图片描述][20201015104802966.png_pic_center]  
完结撒花！

参考：  
[NVIDIA CUDA Toolkit 11.0 安装与卸载(Linux/Ubuntu)][NVIDIA CUDA Toolkit 11.0 _Linux_Ubuntu]  
[pytorch][Link 2]  
[cuDNN Archive][Link 1]  
[CUDA Toolkit 10.2 Download][]  
[Ubuntu18.04下安装pytorch详细步骤][Link 3]

[Nidia]: https://docs.nvidia.com/cuda/cuda-installation-guide-mac-os-x/index.html#:~:text=The%20CUDA%20Driver,%20Toolkit%20and%20Samples%20can%20be,be%20uninstalled%20unless%20the%20--manifest=%3Cuninstall_manifest%3E%20flag%20is%20used.
[NVIDIA CUDA Toolkit 11.0 _Linux_Ubuntu]: https://zhuanlan.zhihu.com/p/122286055
[CUDA Toolkit 10.2 Download]: https://developer.nvidia.com/cuda-10.2-download-archive?target_os=Linux&target_arch=x86_64&target_distro=Ubuntu&target_version=1804&target_type=runfilelocal
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80NTg4NTIzMg_size_16_color_FFFFFF_t_70_pic_center]: /images/20221123/610c379e359d4cebbdab4421253b78ca.png
[Link 1]: https://developer.nvidia.com/rdp/cudnn-archive
[20201015104330286.png_pic_center]: /images/20221123/2613fd52649c4e20a7b1a3d88093c032.png
[20201015104347578.png_pic_center]: /images/20221123/642a3e564a144c5999df6f7be36b8beb.png
[202010151044217.png_pic_center]: /images/20221123/5273049bfd334a5886e96bcd31b255fa.png
[Link 2]: https://pytorch.org/get-started/locally/
[Link 3]: https://blog.csdn.net/weixin_45885232/article/details/107507921
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80NTg4NTIzMg_size_16_color_FFFFFF_t_70_pic_center 1]: /images/20221123/949fc1299528489c81a4985855557b8b.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80NTg4NTIzMg_size_16_color_FFFFFF_t_70_pic_center 2]: /images/20221123/9518163a70f64feebef98e9e4550664d.png
[20201015104802966.png_pic_center]: /images/20221123/01a1f23771d14262b3f82862134a1716.png