I have a python flow in which I can type something and using RAG chatgpt 3.5 it responds.
But now to reduce prompt costs, since a very long role system prompt has to be sent, I am testing with finetunning.
I want to use finetunning so that the 3.5turbo model responds like a virtual assistant and has a name. Additionally I want to use the training to teach information that the assistant should know.
Is this possible?
I have done some tests and although if I repeat some training cases enough it seems to learn the answer and give me similar answers, it seems to ignore the “system” message that I send in each training case.