Find answers from the community

Updated 7 months ago

I have built a Q&A RAG application using

I have built a Q&A RAG application using llamaindex that answers questions based on documents present in a folder. Currently, it always answers questions even if the question is not related to any of the documents present in the folder. In most of these cases, it will make up an answer/hallucinate. Any tips to control this behavior? For example, a user simply typed hello and the app returned some random answer.

6 comments

TTeemu

There are a few ways: you can implement a similarity threshold to make sure that irrelevant documents wont get returned: https://docs.llamaindex.ai/en/stable/module_guides/querying/node_postprocessors/node_postprocessors/?h=similaritypost#similaritypostprocessor

Also which LLM are you using? You can also modify your prompt to help control this

AAnurag Agrawal

@Teemu , I am using llama-2 7B locally. I'll give that a try. What are some other ways to handle this scenario?

AAnurag Agrawal

@Logan M @WhiteFang_Jr Do you have any inputs on this?

TTeemu

With such a small model it will be very prone to hallucinating. Not sure how much prompting would improve it but it's worth a try.

I'd maybe look at the similarity threshold so that those irrelevant documents wont even be included which confuses the model

TTeemu

If you set the similarity threshold correctly, saying 'hello' shouldn't be returning any results

AAnurag Agrawal

Thanks @Teemu!

Add a reply