Large tensor for edge prediction with TF GNN

Hello all,

I try to use TF GNN to predict link in a large graph. The graph tensor has the size of almost 2milions of edges. Then training can run when I reduce below 1 milions of edges but not above.

I use a GPU single node in Databricks with 112 GB and 6 cores.

TF GNN is really intuitive to start with but I’m suprised that we can not use for large graph.

Thankyou very much for your advice,