Hi everyone.
We are building an application which is based on rather complicated prompt that instructs Bot to follow a 6 steps workflow with some function callings. We are going to fine tune GPT3.5 for better function calling performance. I have 2 questions about fine-tune models ’ performance when we update system prompt and about adaptability to personalized system prompt ?
- For example, our fine-tune model is trained with system prompt V1, then if we update system prompt V1 to V2 that could be trivial update such that adding a final step to say goodbye to user at the end of flow. I wonder whether or not fine-tune model with V1 system prompt can work well on V2 system prompt ?
- We are using personalized prompt that including some personal information such as name, address and favorite food, … If we fine-tune model on a set of fix personalized information , then could fine-fune model work well in production when it serves thousand of users by thousand personalized system prompt ?