Can we use a fine tuned model to answer questions? If I have uploaded a jsonl file with the purpose of “fine tune” is it possible to use that same file to generate answers? Or do I have to have to make another file, with essentially the same data, and upload that?
So the idea is that I make an “answers” request, but I choose a fine tuned model I created previously? I guess that could make sense. I created a fine tuned model where the prompts were the titles of articles in a knowledge base, and the completions were the text from the articles themselves. I’m guessing that would not be a useful fine-tune, and I should just follow the answers guide OpenAI API properly.
However, if I were going to use a fine-tune to try and answer questions about a knowledge base, what kind of completions would I put in it?
Your idea makes sense. At the moment the answers endpoint Diane sorry fine tuned models. However you can easily recreate the behaviour on your end using search for retrieval and a fine tuned model for completions.
For example openai-python/answers_with_ft.py at main · openai/openai-python · GitHub