Overview

This project seeks to investigate the possibility of biases in generated answers when using Retrieval Augmented Generation (RAG) techniques with Language Models (LLMs).

Language Models Tested:

• ChatGPT 3.5 Turbo

Embedding Models Tested:

• text-embedding-3-small

Set-up

Follow the steps below to set up the necessary dependencies. LangChain quickstart guide here.

Install Dependencies

pip install -r requirements.txt

Make sure to create your own .env file to store the necessary information for the APIs. The .env file should minimally look like:

API_KEY=insert_api_token_here

Your OpenAPI key can be created here. Make sure to top up the balance to use the api key.

---

Relevant Files

format_output.py is mainly for formatting and visualizing outputs at different stages of the pipeline (loading, splitting, retrieving etc.). The visualized files are under the folder "visualizing_articles".

Note: The reference articles are news articles with various topics (e.g. Covid-19, Climate Change). The articles will be used to in the main file (xlb) to perform RAG on.