.cpu().detach().numpy() vs .data.cpu().numpy()

[alwynmathew]

What is the difference between .cpu().detach().numpy() and .data.cpu().numpy() ?

When should I used one of these over the other?

[albanD]

.data is an old api and should not be used anymore.
So the first one is the right way to go.

[pytorching]

How about .cpu().detach() vs .detach().cpu() ?

[albanD]

The end result is the same.
The second one is going to be imperceptibly faster because you don’t track the gradients for the cpu() op. But nothing else.

[pytorching]

Thank you very much. That answers my question exactly.
BTW, I don’t know if it’s allowed or forbid. May I ask you to look at another question I post?

nn.DataParallel: TypeError: expected sequence object with len >= 0 or a single integer

In my forward function: def __call__(self, train=True): if train: predicted = self.forward(...) loss = .... return loss # return a single value that's fine # loss.size() = the number of my GPUs. else: predicted = self.forward(...) return predicted # expected sequence object with len >= 0 or a single integer # In validation step I want to return the whole predict labels for other purpose # predicted with shape [16, 1] on eac…