The model is actually saved in a serialized format, so there’s no conversion that needs to happen. TensorFlow does have a C++ API, so that if we want to run the model from a C++ program in production, that is possible.