Derive the gradients of y_hat w.r.t the inputs to an one-hidden-layer neural network

The one-hidden-layer neural network looks like below.

How would you derive y_hat’s gradients with respect to the input X?