Gradient failure in torch.nn.parallel.DistributedDataParallel

[mrshenli (Shen Li)]

I see. In that case, is it OK to just ignore the above warning?