Hi,
My full graph object is too large for the GPU. Since I got two GPUs, I wonder if I could split the graph into two. I did using:
dgl.distributed.partition_graph(g,graph_name=‘test’,out_path=’…’, num_parts=2, shuffle=True)
but I cannot get the part for loading, as I thought loading it would get my 1 out of the 2 dgl graph object.
graph_parts_0 = dgl.distributed.load_partition(’…/test.json’, 0) but this is tuple of values, graph_parts_0[0] is the schema, graph_parts_0[1,2] are node and edge features, with altered names, and they do not match the dimension reported in graph_parts_0[0]. So I am confused, as I thought I would get two graphs that would be otherwise identify in terms of structure like the original g, so that I could use for training. But this is not the case. Any advise would be grateful!