For a 1-feature linear model pred = w * x + b and the mean-squared-error loss against targets y, use autograd to compute the gradient of the loss with respect to w and b. Return a length-2 tensor [dL/dw, dL/db].

Input

x = tensor([1., 2., 3., 4.])
y = tensor([2.0000, 4.1000, 5.9000, 8.2000])
w = tensor(1.)
b = tensor(0.)

Output

tensor([-15.3500,  -5.1000])