Anonymized version of the repo for my MSc thesis: "Interfacing Operating Systems with Large Language Models"
- Python:
3.11.2
python3.11 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
- Make sure
OPENAI_API_KEY
environment variable is set.
- Alias the script in your
.bashrc
/.zshrc
like this:alias roger="noglob [path_to_venv]/venv/bin/python [path_to_repo]/thesis/src/agent/roger.py"
- Also add the repository to your
PYTHONPATH
export PYTHONPATH=$PYTHONPATH:[path_to_repo]
roger [Natural Language Pompt]
roger
's implementation lives here, you can edit the setup and fewshot promptsin utils.py
.
Feel free to change model="gpt-3.5-turbo"
to model="gpt-4"
in roger.py
to use GPT-4.
The NL2Bash (Lin et al. 2018) dataset is used to fine-tune various GPT-3 models here. Data preparation, fine-tuning, and shell scripts to get predictions from the fine-tuned models can also be found here.
Contains an inference cost estimation script.
This directory contains the user study material. There are also scripts to annotate the experiments.
The dockerStuff
directory was used to set up a container for participants to use in the second task of the user study.