Local Deployment of Minimal SQuARE

This branch hosts a minimal version of SQuARE for the convenience of local deployment.

To get started, please run:

pip install -r local_deploy/requirements.txt
bash install-local.sh

to generate a docker-compose.yaml file (and a authentication token for local deployment in ./local_deploy)

And then:

docker-compose up -d
python local_deploy/embed_query.py  # An example of calling model API to embed a query
python local_deploy/deploy_ds.py  # An example of deploying a new datastore

This will start the minimal SQUARE service and run the example script local_deploy/deploy_ds.py to - Download example corpus scifact; - Upload the corpus (to the ES in SQUARE); - Do BM25 search; - Embed and index the corpus with DPR (GPU 1min, CPU 30min); - Start a Faiss container to load the built dense index; - Do dense search.

One can also get access to the SWAGGER UI page for the interactive API of Datastores. To do so, one can do

export SQUARE_PRIVATE_KEY_FILE="local_deploy/private_key.pem"
square_token  # A Python entry point from `square_auth`

to get the token for authorization. Thereafter, just open http://localhost:7000/docs# and type in the token by clicking on "Authorize". Then you are ready to go.