In my model, the input is a graph data in the form of edge-index and the node features. After a few iterations of training on graph data, loss (EDI
