Fine tuning with knowledge base

I am working on developing conversational chatbots that exhibit a specific style in their responses. While I’ve successfully fine-tuned models to achieve this style, I’m now looking to integrate a separate knowledge base into the system. The goal is for the chatbots to refer to this knowledge base when users ask questions that match its data.

I’ve tried assistants with retrieval files, but it didn’t seem to work too well. The knowledge base wasn’t consistently being used, it often answered using the model alone.

How can I effectively integrate a knowledge base into my chatbot system without incorporating its data into the fine-tuned model?

You have also tried GPTs?

1 Like

Besides @SomebodySysop suggestion, you could also consider building your own RAG system using embeddings. This would provide you with more control than in the Assistant‘s case.