Chat Completion with Vector database

API
1

Hi there,
Currently I have a system where I use an assistant connected to a vector database with over 50 files. However the problem is i don’t want it to be a chat system like an assistant rather I give it a prompt and it will return that result.
Just like the chat completion.

I am wondering can I create a system like this

let completion = await openai.chat.completions.create({
  model: 'gpt-3.5-turbo-0125',
  // max_tokens: 4000,
  messages: messages,
  temperature: 0,
  functions: functions
});

const messages = [
  {
    role: 'system',
    content: 'Generate questions and answers in the form on an array'
  },
];

But I use a vector db with over 50 files.

The reason why I am doing this is it currently works if I just have 1 file, because it does not go over the token limit. However I have over 200 files and I can not go though all of them and generate the questions because:

  1. I don’t want the questions to be the same
  2. I have more requests that I need and I don’t want to send over a 1000 requests just for 200 files.

Thank you

2

Welcome to the community!

Is this all you’re sending? It would need to know what source of data it should make the questions from…

Or maybe I’m misunderstanding your question?

3

Sorry Paul,
I tried to generalise the prompt because it’s a bit long with a few contraints.

What I am trying to do is?
I have 50+ documents and rather than having an assistant with a file search feature, go through all of the 50 documents (pdf word txt…) and run a series of prompts eg
Make 50 questions or summarise it.
However if I use an assistant it gives it more to me as a conversation so eg

Here are your questions

  1. …
    With styling

But I just want a strict array eg
[{question: “…”},…]
Just like the chat completion function does.

Hope this clears it up
Thank you