How to customize a conda environment

To perform meaningful calculations, we have to install additional packages in our conda environment. If you don’t have a conda environment installed, refer to: How to create and activate a conda environment. Here, we restrict ourselves to a simple example. The code we are going to work with is /home/ENV-FILES/user-docu-examples/PyTorch-1_x-MNIST.ipynb. To use this code, copy it to your home directory, e.g.,

1. Open a terminal and type:

    cp /home/ENV-FILES/user-docu-examples/PyTorch-1_x-MNIST.ipynb /home/<username>
   

   If your are interested in other examples, refer to the Advanced options section.

2. Once the file is in your home/<username> directory, you can open it from this location and modify it at will.
3. First, check your GPU driver version to know your CUDA limitations.
4. In this example, the Torch library and the Tochvision package are required. Compatible versions are listed in https://download.pytorch.org/whl/torch_stable.html.
5. Install PyTorch 1.10.1 and Torchvision 0.11.2. To do so, activate the conda environment that you want to work with and then install the packages in it:

    conda activate <name>
    conda install pytorch==1.10.1 torchvision==0.11.2 -c pytorch -y
   

6. Once the installation is completed you can list all your packages considering

    conda list 
   

7. Also, you can list specific packages, e.g.,

    conda list cuda
   

8. As you can see in Fig. 1, when you installed PyTorch 1.10.1, cudatoolkit 11.3.1 was also installed.

   

   Fig .1.

9. Your environment now has the necessary packages to run PyTorch-1_x-MNIST.ipynb. To run it, refer to: How to run a Jupyter notebook file.