Lack of gradient when creating tensor from numpy

Question

Can someone please explain to me the following behavior?

import torch
import numpy as np

z = torch.tensor(np.array([1., 1.]), requires_grad=True).float()

de