bleak cairn
Hey everyone! I've read the docs and searched the forms for questions regarding fine tuning and function calling but haven't found exactly what I'm looking for.
I know when fine tuning you want the input to match as close as possible to expected input to the model
And I know that you can use fine tuning to ensure structured data.
But we have a mechanism in functions/tools for that already. If I'm going to be using structured output (I'll be using it with structured output parsers with zod schemas and langchain not that it matters for this question) should I still provide my function in my training data?
I would assume the answer is yes because we still want the input (whether it be user, assistant or otherwise) to match the same as when the model is actually in use.