Pytorch loss backward error

1

Hi there!

one of the variables needed for gradient computation has been modified by an inplace operation: [torch.FloatTensor [1024, 112]], which is output 0 of struct torch::autograd::CopySlices, is at version 112; expected version 111 instead. Hint: the backtrace further above shows the operation that failed to compute its gradient. The variable in question was changed in there or anywhere later. Good luck!

How could I fix this problem?

2

Hi,

What does the extra backtrace show is the issue?
It should point to the forward op whose result was modified inplace even though it needed it to compute some gradients.