Hi folks.
Would appreciate some advice on the best way to fine-tune a model where I’m looking to understand multiple parameters.
Most of the “out of the box” fine-tuning formats are as follows:
- role:user “What color is the sky?” role:system: “blue”
- role:user “What color is the ground?” role:system: “brown”
and I can fine-tune to understand expected answers…
There are also multi-turn chats like the one below from OpenAI.
{“messages”: [{“role”: “system”, “content”: “Marv is a factual chatbot that is also sarcastic.”}, {“role”: “user”, “content”: “What’s the capital of France?”}, {“role”: “assistant”, “content”: “Paris”, “weight”: 0}, {“role”: “user”, “content”: “Can you be more sarcastic?”}, {“role”: “assistant”, “content”: “Paris, as if everyone doesn’t know that already.”, “weight”: 1}]}
My question: How do I fine-tune if I want to understand the result of the conversation, or better yet…multiple results.
For example, what if I had the following data like category and outcome?
role:user {“What color is the sky?” role:system: “blue” category: “question” outcome_of_convo: "answered}.
I have tens of thousands of chats with things like summaries or status that I want to capture as well as the convo itself. Ideally, I’m looking for non-dev friendly tools that multiple people at my company could use.