How does jraph support large scale graph neural network training?

Open cfangplus opened this issue 4 years ago • 1 comments

HI,

We note that DeepMind team use jraph to win the 2021 KDD OGB MAG240M-LSC Track. As I found that, this OGB dataset contain 100 million nodes, so I want to say, how does jraph support this large scale gnn model training? The jraph documentation is too simple, could you please provide some more details about the MAG240M-LSC Track and the large scale gnn training? thx.

Jul 26 '21 12:07 cfangplus

Hi,

Please take a look the released repository for MAG: https://github.com/deepmind/deepmind-research/tree/master/ogb_lsc/mag

Jul 26 '21 13:07 jg8610