The trained model is a guarded treasure. There's a reason why very few ML papers publish their trained models; or Deepmind/OpenAI never releasing the resulting models - it's expensive to train the network, but cheap to use it afterwards. You don't want anyone unauthorized to get their hands on it.