How Pytorch calculate Gradient in linear regression loss function

Question

I have been following this tutorial on PyTorch linear regression. the loss function is torch.sum(diff * diff) / diff.numel() where diff is Target - predicted va