Configuring pytorch in win10 environment

utilize Anaconda To configure pytorch.

One , To configure cuda as well as cudnn（gpu Speed up ）

（ If not nvidia Video card , Only use CPU, Ignore this step ）

Download it on the official website cuda as well as cudnn after , Start configuration cuda Environmental Science .

Reference blog ：win10 Next CUDA and CUDNN Installation （ Hyperdetail ）！ Close test effectively ！_ Nobody likes a person's blog -CSDN Blog _ install cuda and cudnn

Two , install Anaconda

2.1 Download and install from the official website  Anaconda. 

Open from the start menu Anaconda Medium Anaconda Prompt, appear （base）, Successful installation , as follows ：

（ Configuration failed. Please refer to the pit stepping record at the end of the article ）

2.2 Configuration environment

After successful installation , Stay on top Anaconda Prompt On the page , Input  

conda create -n pytorch python=3.8

  Then input y. Install the required packages .

next , We type in ：

conda activate pytorch

  You can see

The leftmost base, Turned into pytorch. It's in pytorch Compiler environment for .

2.3 pytorch Installation

stay pytorch Select our corresponding environment in the corresponding part of the official website .

Pytorch Build： General choice Stable（ Stable version ）.

Your OS：PC Operating system of .

Package： We are in anaconda Installation on , So choose Conda.

Language： choice Python.

Compute Platform： Here we need to configure according to the cuda To choose from .

win+r, Input cmd Call up the command line window . Input

nvcc -V
nvidia-smi

Can view the current cuda edition .

In the bottom line （Run this Command） Then copy the corresponding command .

For the above figure, i.e ：

conda install pytorch torchvision torchaudio cudatoolkit=11.1 -c pytorch -c conda-forge

When it's done , We are Anaconda Powershell prompt Copy the above command in the window .

Then input y. Wait for the installation to succeed .

（ Slow speed 、 See pit record for failure ）

2.4 Verify environment installation  

      close Anaconda Powershell prompt, Open it again , Again by order conda activate pytorch Get into pytorch Environmental Science .

Input in sequence ：

python

 

import torch

torch.cuda.is_available()

  so , return Ture, explain cuda Installation successful , otherwise pytorch Only use cpu function .

3、 ... and , Record on pit

3.1 Above, 2.1 Partial tread pit

  If a similar error is reported here ：

  Please check the system environment variable ,Path route . All variables must be written in sections , Out-of-service “;” Write it in one piece . 

3.2 Above, 2.3 Installation failure encountered in 、 The speed is too slow .

（1） After reading many blogs, I suggest using Tsinghua mirror , But I tried many times and failed . For the first time, it was quickly installed CPU Version of , Cause the computer to reinstall and start all over again . It is recommended to download it on the official website , I put it at night , It will be installed in the morning . Try a few times more .

（2） Do not hang up during installation VPN, Inevitable failure , Or report a mistake . I wasted an afternoon because of this .

（3） to update conda command .

3.3 torch.cuda.is_available() return false Probable cause

（1） Check if there is GPU.

（2） Check CUDA as well as CUDNN Environmental Science 、 System Path Variable configuration .

（3） Check if the graphics card driver is up to date .