Typo instantiating loss at zero

[aymebou]

tutorials/intermediate_source/char_rnn_generation_tutorial.py

Line 281 in 48e5ccc

loss = 0

Hello folks at Pytorch,

When instantiating loss at a real zero, i think you rather meant something like (i don't know if there is a better way but I usually do it this way) :

loss = torch.autograd.Variable(torch.Tensor([0]))

[zabboud]

I would take this on but first I wanted some clarity on the question - is the question whether setting loss = 0 vs. loss = torch.autograd.Variable(torch.Tensor([0])) is better? Or just to explain why loss = 0 is used instead?

[abhi-glitchhg]

When instantiating loss at a real zero, i think you rather meant something like (i don't know if there is a better way, but I usually do it this way) :

I think instantiating the loss with 0 (or python int dtype) is not a problem because once you use the add operation with an integer and a tensor

tutorials/intermediate_source/char_rnn_generation_tutorial.py

Line 286 in 48e5ccc

loss += l

the dtype is promoted to tensor[1][2].

But initializing it as a tensor won't do any harm and could avoid future confusion like this.

[akjalok]

/assigntome