Hi, My job is to fine-tune a gpt3.5 model. I have two datasets. In first dataset I have prompt’s information including prompt, prompt desc, prompt title, and category. I want that somehow my model know these data. In my second dataset, I have questions regarding these prompt info and target prompt title. I want to finetune the model such that if user asks a question to it, it should suggest the most suitable prompt title to the user and if their is no such prompt title it should say accordingly that no prompt title matches your requirements.
So as a solutions I created a single example (example_0) from first dataset which contains all prompts info. and the second dataset which contains questions and answers I divided it in train and test set and added example_0 at first position in train set and performed fine-tuning. But the problem is for some cases it reframes the label (prompt_title) or for new data it generates the prompt title but rather it should return “No matching prompt title” .