This is the error I have encountered RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.FloatTensor [20, 80]], which is output 0 of TBackward, is at version 2; expected version 1 instead. Hint: the backtrace further above shows the o…

You would need to recompute the forward pass using the generator and its already updated parameters to create the new forward activations. Alternatively, you could also delay the optimizerG.step() call until both backward passes calculated and accumulated the gradients using the same (initial) param…

Struggling with a Runtime Error related to in-place operations

Alexey_Demyanchuk (Alexey Demyanchuk) November 13, 2020, 8:35am 4

Could this be a potential clue?