README.md

Janus: A Project Mercury Endeavor

An interactive interface for GPT-X, built with Streamlit.

Setup on Stanford NLP Cluster

# Get a GPU machine: Titan-X is known to work while K40 does not
nlprun -a janus -g 1 -s bash

# Clone the repository
git clone https://github.com/stanford-mercury/janus.git
cd janus

# Use the existing conda environment available on the cluster
conda activate janus

# Alternately, create a fresh conda environment
conda create -n [env_name]
conda install pip
pip install -r requirements.txt 

Setup on MacOS (CPU-Only)

Clone the repository, and create a conda environment with environment-osx.yaml.

git clone https://github.com/stanford-mercury/janus.git
cd janus
conda env create -f environment-osx.yml

Running Janus

Run the streamlit application and you're good to begin!

# Auto-selects device
streamlit run main.py

# Use CPU-only
streamlit run main.py -- --device 'cpu' 

Navigate to the URL to open the app.

Note (Stanford NLP Cluster): GPU tested on titanx. GPU k40 doesn't work due to torch+cuda version issues, launch the app in CPU-only mode.

First-Time User

You'll have to register the first time you use the app. The master password is mercury-bagel.

Data Collection Notice

By default, Janus collects and stores all your interactions and generations in the app (whether you save them explicitly or not). When running the app locally, this data is stored under data/<username> and is not transmitted at this time.

Issues

If you have any problems using Janus, please file a Github Issue. Feedback can be given directly to the Project Mercury team.

Contributing

If contributing to this repository, please make sure to do the following:

• On installing new dependencies (via pip or conda), please make sure to update the environment.yml files via the following command (note that you need to separately create the environment-osx.yml file by exporting from Mac OS!):

  rm environment.yml; conda env export --no-builds | grep -v "^prefix: " > environment.yml

---