Back propagation in pytorch to calculate component wise derivatives of vector

Question

Suppose I have a vector valued neural network f which takes in a scalar x, i.e. f(x) = vec{y} = [y_1, ..., v_n].

I want to use automatic differentiation in Pytorch to calculate [y_1', ..., y_n'], where $y_i'$ is the derivative of the ith component of f(x) with respect to x.

I'm pretty stuck.