The following are the variables that need to be set:
GIN_MODE: In which mode Gin should be running (can bereleaseordebug)MODELS_HOST: The hostname of the DL models container (usemodelsif in release mode,127.0.0.1if in debug mode)
To download the Universal Sentence Encoder (USE) model, run the python script in scripts/download-use.py by running the following commands (you should run these commands while in the backend directory):
conda env create --file=environment.yml
conda run -n api-scout python ./scripts/download-use.pyYou will now have a new directory in models called universal-encoder. This model will be used by the docker-compose.yml file to serve the model in a container.
For replication purposes, in this repo you will find both a Dockerfile and a docker-compose.yml file. The Dockerfile will create a Docker image with a build of the Golang backend in it. To create the image, simply run:
docker build -f Dockerfile -t api-scout-backend:latest .Once the image and its dependencies have been downloaded, you can now spin up the backend containers. To do that, simply run:
docker-compose up -dThis will create a Docker container for the Golang backend and for the USE model. The backend will now be able to make HTTP calls to the USE model to embed queries and documents. N.B.: The USE container will not expose any ports, it will be called locally by the backend by means of the be-network shared network.
For the backend to work, both an ElasticSearch instance and a MongoDB instance should be up and running.
To generate and consult the documentation, you can use the following commands.
To generate the documentation, first you need to make sure that you have all the necessary dependencies installed. Run the following commands:
go install go.abhg.dev/doc2go@latest
npm install -g pagefind@latestOnce all dependencies have been installed, run:
doc2go -config ./app/internal/doc2go.rc ./app/internal/...To consult the documentation of the app/internal API, run the following command:
cd docs && python -m http.server 8000