It seems that torch.rand generates [0,1) uniformly based on the official documentation instead of standard Gaussian. Is this intended? Thanks