Difference between fine-tuning and assistant retrieval

I had a question running, what if we don’t turn on the plugin of retrieval in assistant and still upload a knowledge base file, what will happen.
and second question is what is difference between fine-tuning a dataset and giving a dataset to the assistant retrieval

Hi -

To your second point:

Fine-tuning serves to adjust model behavior, enabling it for example to perform tasks in a certain way or produce output in a certain style. It does not serve to teach a model new knowledge.

Providing a dataset for the assistant retrieval, on the opposite, is equivalent to injecting knowledge, which can subsequently be retrieved and incorporated when responding to user requests.


so fine-tuning is just like giving context on how to behave to questions right??
can we achieve this in assistants to give a certain style behaviour ??
( EDIT: i got the solution, we can give custom instructions to the assistant or thread to behave and maintain context )


yep - you got it right :slight_smile:


1 Like

When you upload a knowledge base (KB) file to the Assistant but do not enable the retrieval plugin, the Assistant will not automatically use the uploaded knowledge base for generating responses. The retrieval plugin is what allows the Assistant to access and use external data (such as a knowledge base) to inform its responses. Without this plugin being active, the Assistant will rely solely on its pre-trained knowledge up to its last training cut-off and any fine-tuning data it has been trained on. Essentially, the uploaded KB file remains unused and does not influence the Assistant’s responses.

1 Like

With this reply @RobSteele , A new concern came into mind, since we can’t update the uploaded knowledge base files, can we use an external file as a knowledge base for the assistant or a thread or message with function calling tool ??