Gradient descent (not Neural Network) in PyTorch c++ without derivative

Question

I have a function f(X) which takes a 10 element array/tensor and returns a float. I am looking to find X to minimise f(X). f(X) >= 0 for all X, and I cannot calculate t