is it possible an old-fashioned batch version will be developed

[tiger-tiger]

Is it possible there is an option, we can just use [B, F, N] for nodes, [B, F, E, E] for adj matrix as inputs, pyG automatically converts them to some underlying data format.

meanwhile we generate batched graphs with nodes [B, F,N] & adj [B, F, E, E] of same size, so loss can be easily calculated for labeled graphs.

[rusty1s]

We might be able to integrate something like that once NestedTensor is in a more stable state. No promise though :)