I wanted to do a forward pass for a GNN (or Tree NN). I know that for vision one can process the entire batch in one go by having the batch be a dimension in the tensor. Is something like this possible for GNNs (Graph Neural Networks)? I am worried that having to do a forward pass for each example for my GNN might slow things down, but perhaps there is no other way.
I assume multithreading might be possible but I don’t know how that would work with multiple threads/processing using the same gpu…plus debugging sounds like a nightmare. But I am curious what is the standard practice for this since GNN research exists.