A bit of context, I’m working on designing several different “generators” that use the OpenAi API to create content based on parameters given by the user. It’s always a single response from the assistant (not a back and forth chat). For example, one generator is “Email Generator” where you can fill in certain fields, those input field variables are put into different parts of a prefabricated prompt, send to the API and it will output a generated email. I always output in HTML surrounded by triple backticks so my platform can easily interact with the output.
Question:
If I have to create 20 of these generators, all doing different things, what are the pros and cons to fine-tuning 20 different models with 100 examples for each generator, vs fine-tuning one model with 2000 examples (100 from each generator type)?
I’ve been leaning towards a different model for each, but wanted to know if anyone had an opinion one way or the other, if there were any limitations, etc.