PyTorch

Overview

 PyTorch  is the main deep learning library that we use in the group.  PyTorch Lightning  helps reduce boilerplate in PyTorch code. There are also companion packages, like  PyTorch Geometric  and  dgl , for graph-based deep learning models.

Installation

To install PyTorch, go to the  getting started page . Select the suitable version to install based on your given operating system and environment.
If you are installing PyTorch on your local machine and your local machine has an NVIDIA GPU, you will likely need to install a compatible CUDA version. For instance, consider the image below. In this scenario, I would need to install CUDA 12.6 on Windows first. The easiest way to do this is just by Googling "CUDA Toolkit 12.6 Downloads" and clicking the link from NVIDIA. Note that the most recent versions of CUDA may not be supported on the most recent version of PyTorch, so make sure they are compatible.
To confirm that your PyTorch build can use the GPU, run the following. It should return True (provided that you are running it on a machine with a GPU). If using an HPC cluster, make sure you run this from a GPU login node or compute node.
import torch

print(torch.accelerator.is_available())