strange behavior

hi, I am using your wonderful plugin, however I have some issues with fine tuning. I have created a lot of question/answer like in your example, around 200. I followed your guide, step by step. But when I use the trained model the behavior is ver strange. Whatever I use in prompt, the chatbot reply is not a straight answer, but a dialogue, I mean the answer is formatted like: (AI: “answer” Human: “another question” AI: “another answer” Human: “another question” … ). Every answer contains at least 5 or 6 question/answer in form of a dialogue, usually it repeats the same q/a multiple time. And most of all this dialogue rarely contains the text I have used for tuning. That’s very strange.

BTW: what I want to achieve is a bot trained on the themes (and tones) of a book, so basically I have tried to create a bunch of questions related to these themes, using the text of the book (tweaking it a little bit) as answer. The idea is that the reader of the book should be able to “discuss” about the themes with the chatbot as if it were “the book” (or the author).