Somewhere in my loss function, I invert a complex matrix of size 64*64. Although complex matrix inversion is supported for torch.tensor, the gradient cannot be computed in t