When I load dgl.data.GINDataset, I find that graph.ndata[‘attr’] is a th.float64 Tensor instead of float32. However, the default dtype of a pytroch Tensor is float32(the weight matrix of a nn.Linear module is also a float32 Tensor). So we will meet Tensor type conflict when forwarding a model.
This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.