How can I train GPT-2 with ecological signals and generate pseudo-ecological signals?

The GPT-2 model first performs 1,000 training steps for each of the 20 channels of EEG data in each class. There is a paper that says to then generate synthetic data for n classes using n GPT-2s. However, I don’t understand how to do this. Is it possible for someone to teach me since the method of generation is not written in the paper?