This repository allows you to recreate the results of this blogpost.
This code has been tested with with Python Python 3.9.13.
- Activate your virtual env
- Install Pytorch via the installation instructions given here: https://pytorch.org/get-started/locally/
pip install -r requirements.txt
You should now be able to run python chat.py and see the code build a vector index from the contents of data/zef.txt, then drop you into command-line chat loop with the Zef chatbot. By default the script makes calls to the OpenAI API, so you'll need to set the OPENAI_API_KEY variable in your shell environment: https://help.openai.com/en/articles/4936850-where-do-i-find-my-api-key
chat.py accepts an -m flag that can be either an OpenAI GPT model or a model on Huggingface, i.e mistralai/Mistral-7B-v0.1
You can bring your own data by pasting text chunks seperated by the string \n-----\n into a single document, similar to the layout of data/zef.txt. Then make a config document of the format /configs/zef.json and pass it to chat.py via the -c flag.