How I bypassed PyTorch OOM errors with a Zero-Copy C++ Graph Engine

If you have ever tried to train a Graph Neural Network (GNN) on a massive dataset, you already know...