I got the same problem when using torch.fft.fft().
As an alternative, I used the old representation, torch.fft(). This function can avoid this bug.
I got the same problem when using torch.fft.fft().
As an alternative, I used the old representation, torch.fft(). This function can avoid this bug.