Today,I find that installing Pytorch using conda is very slow. Thus I recorded the solution of this installation problem.
1. First, you should clearly know the specific name of installation package.
you can use the normal way to obtain the name of package. For example, if using the following instrction to install pytorch1.0
conda install pytorch==1.0.1 torchvision==0.2.2 cudatoolkit=10.0 -c pytorch
It will be very slow when install pytorch and cudatoolkit packages. But, you can obtain the specific version of these packages. Like this:
Then, you can find the name of pytorch package is '1.0.1-py3.6_cuda10.0.130_cudnn7.4.2_2'
2. Then, download the package after obtaining the package name.
you can download the package from Tsinghua Open Source Mirror website (https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/pytorch/linux-64/?C=M&O=A). After downloaded, using the 'conda info' to show dir path the package cache of anancoda [Here, the dir is /home/xiongdeng/ananconda3/pkgs]. Like this:
And after that, move the pytorch package to this package cache dir.
3. Next, Install PyTorch package.
Using ‘conda install [path]/pytorch-1.0.1-py3.6_cuda10.0.130_cudnn7.4.2_2.tar.bz2’ instruction to install it