I am trying to use R-GCN for a link prediction task.
I looked at the documentation on https://docs.dgl.ai/tutorials/models/1_gnn/4_rgcn.html?highlight=link%20prediction
It does not appear that the current implementation supports GPU usage? Is there an easy way to adapt the implementation to support GPU usage and even multi-GPUs?
You can check our example code where we do use gpu for training. To use multi-GPU, see this issue.
Got it thank you :). For some reason, I missed the gpu part in that code.
I am trying the link prediction example on my workstation, with has a GPU with 4GB of memory. I am running out of memory when running the example. Do you know what the memory requirement for that example is on a single GPU?
I think it should be 12GB or 16GB.