Our goal was to create a chatbot capable of answering questions related to AI policy. To achieve this, we explored various approaches and selected the most effective one:
- LLaMA-7B and Chat (Pre-trained/vanilla)
- LLaMA-7B and Chat (Prompt-engineered)
- LLaMA-7B (Fine-tuned)
This repository contains the following files and directories:
final-presentation.pdf: A brief presentation about project goals, approaches, and results.notebooks: A folder containing all Jupyter notebooks, which include:data-cleaning.ipynb: Code to prepare data for prompt-engineering and/or fine-tuning.prompt-engineering.ipynb: Code to run prompt-engineering.fine-tuning.ipynb: Code to run LoRA/QLoRA fine-tuning.evaluation-viz.ipynb: Code to evaluate the predictions and visualize the results.human-evaluation.xlsx: Excel table with the results of human evaluation.
demo.mp4: A video demonstrating the performance of prompt-engineered and fine-tuned models. Also available on Youtube: https://youtu.be/dnIPv0LCaZwfinal-report.pdf: A detailed report in PDF format summarizing the approach, methodology, results, and insights from the analysis.README.md: This file provides an overview of the project, its objectives, and the contents of the repository.
See the full list of references in final-report.pdf.
Note: All data used in this project is sourced ethically, and the analysis adheres to the highest standards of research integrity and ethical guidelines.