Hongzhi (Steve), Chen comments

Results 53 comments of


                                            Hongzhi (Steve), Chen

DistDGL doesn't create the mapping between etype/ntype and etype/ntype IDs correctly.

Hi Da, can you help us clarify the reproduce steps? We are having a hard time guessing which piece went wrong :(

Support state save/load for DGL sparse optimizers

Duplicated with https://github.com/dmlc/dgl/issues/4744?

Support state save/load for DGL sparse optimizers

confirmed dupilication.

[Performance] GSpMM regression with FP16

Hi @yaox12 , is this done already?

[GraphBolt][Examples] Folder Location

Sure, feel free to do so.

End2end test of distributed GraphSAGE + ogbn-papers100M

This has been done by Da's team.

End2end test of distributed RGCN + ogbn-mag240M

This has been done by Da's team.

[GraphBolt] TorchData Pytorch support

The way we implement DataLoader (https://github.com/dmlc/dgl/blob/658b2086b09bbd76c3d3f488af2b155a1c921052/python/dgl/graphbolt/dataloader.py#L79C7-L79C17) right now isn't perfect. It makes a lot of assumption that might cause problems later. Once those problems hit, we should redesign it. We...

[Graphbolt] Optimize exclude edges with more than int32 number of edges.

The primary optimization has done. For most cases, which has less than int32 number of nodes, we have the optimization in place. The future work is to support cases that...

Sparse Ops: HGNN example

duped with https://github.com/dmlc/dgl/issues/4870