Multiple GPU about dctts-pytorch

[chaiyujin]

@Tywors
Insert graph = nn.DataParallel(graph) at

dctts-pytorch/pkg/train.py

Line 127 in b69b9f6

"graph" : graph.moduel.state_dict() at

dctts-pytorch/pkg/train.py

Line 62 in b69b9f6

"graph": graph.state_dict(),

[chaiyujin]

@Tywors
Actually, after graph = nn.DataParallel(graph), the original graph is wrapped by a DataParallel. It's possible to access original graph by graph.module.
Maybe you have to change graph to graph.module at several places.
I''m sorry that I don't have time to help you do all changes, because I'm focusing on my private project.

[Tywors]

Thanks for your answer!

With this change, this is the result in train:

File "main.py", line 62, in <module> main() File "main.py", line 47, in main train(args.module, args.load) File "/content/dctts-pytorch/pkg/train.py", line 54, in train train_text2mel(load_trained) File "/content/dctts-pytorch/pkg/train.py", line 177, in train_text2mel atten_mask = torch.ones_like(graph.attention) File "/usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py", line 518, in __getattr__ type(self).__name__, name)) AttributeError: 'DataParallel' object has no attribute 'attention'

any suggestion?

[Tywors]

Thanks for your time!
I continue trying!

[lxy444]

Thanks for your time!
I continue trying!

Have you solved the problem?