Although doing RAG does it worth fine tuning the LLM on the documents? - Llama2

Hi everyone, I got a question.
I want to do a chatbot specialized on a specific topic. For this I got plenty and plenty of documents. In a wikipedia style.
Obviously I am going to do RAG (Retrieval Augmentation Generation) to give the LLM context (the most relevant documents so) to answer the user’s question. But my point is does it worth fine tuning the LLM on the raw document ?
I can’t find any information about that.
Can someone enlight me ?

If the chatbot uses domain specific terminology or nuances then fine-tuning can help the model understand and generate content that aligns better with your domain.
However you might want to start by experimenting with pretrained Llama2 model and evaluate its performance. If you find that it falls short of your expectations, then consider fine-tuning as an option.

1 Like