An Adversarial Process of Two Pre-trained Models for Natural Language Generation - View it on GitHub