How to turn trained checkpoints into C ++ programs

verifier = ppo.PPOTrainer(config=config)
action = verifier.compute_action(obs)

after trained, i have a checkpoint, it include a policy what i want.i use it like above.
but restore a checkpoint will cost me 50 seconds and make debug slowly and it only work in python now.
i know the policy is just a function, i pass a n dimension array, and return a n dimension array.
can i find a way to use it like overwrite it by c++?just like call a c++ function, more quick and light.

int compute_action(vector<double> param)
    return 0