Vectors and assistants

[midnight surge]

Hello,

I have a question about how vectors work with assistants. I would like to summarize PDFs by passing sequences of instructions to an assistant. For now, I have created a vector from my PDF and used a generic prompt to summarize the important points of the document.

However, I wonder if this is the right approach because I think that by using a vector, the API internally uses RAG (Retrieval-Augmented Generation), and since I am not asking specific questions, it may not be able to summarize the document optimally. Is this really how it works, or am I using the right method according to you?

Thank you!

[astral mauve]

https://platform.openai.com/docs/assistants/tools/file-search/how-it-works

[midnight surge]

OK, thank you, so it really is RAG.
What would be the best approach in this case? Should I fine-tune the same model every time I have a new document or split the document into chunk and summarize each chunk or something else ? 🤔

[astral mauve]

fine-tuning is not really for adding new info. if you cant get it to work with the assistant chunking and other settings you might just need to do your own file parser and summarizer

[midnight surge]

Yes that’s true
So i parse the file with my own parser ans summarise each part with GPT ?

[gloomy hornet]

why do you want to use vectors?

[midnight surge]

i used vector because it don't find any other way to pass a file to openai to analyse it but it seem to be the wrong way

[copper raft]

If your PDFs are small enough then just convert them into text and pass that text into the context to the chat completion API. If they are too big split them into chunks and summarize each chunk separately , and then summarize the collection of all the previous summaries.