GraphDataLoader with SAINTSampler

When I load a set of graphs with GraphDataLoader during model training (using PyTorch Lightning), I want to perform graph sampling with SAINTSampler (1) on the loaded batch, or (2) sampling on each graph before being batched.

How can I adapt the GraphDataLoader for SAINTSampler?

(Note: Due to the size of the graphs, they cannot be batched into a single graph by dgl.batch, and use dgl.dataloading.DataLoader afterwards.)