I want to FineTune chatbot that needs to answer questions as truthfully as possible using provided context via Embeddings.
Example of prompt:
Answer the question as truthfully as possible using the provided context, and if the answer is not contained within the text below, say "Sorry, cannot help with that".
Context: <>
Customer: How is this product turned on?
Ai: You have a little switch on the back!
Context can of course change depending on what is the question and what embedding returned for that question.
When fine tuning:
- Should fine tuning data consist of Context and answers? In my case bolded text.
{“prompt”:" Answer the question as truthfully as possible using the provided context, and if the answer is not contained within the text below, say … “completiton”: " You have a little switch on the back!\n"}
- Or is it enough to use only the Questions and Answers section?
{“prompt”:" How is this product turned on?“,“completion”:” “completiton”: " You have a little switch on the back!\n"}