Deploying

Docker

Building

Build the Docker image:

$ ./build.sh

Deploying

Run with docker-compose:

$ docker-compose up -d

or manually:

$ docker run --name marcell-sl-pipeline -d \
    -p 127.0.0.1:5000:80 \
    marcell-sl-pipeline:latest

GPU support

To enable Docker GPU support on your host, please refer to the Nvidia docs. Once you have nvidia-container-runtime set up, you can add the folowing runtime definition to /etc/docker/daemon.json:

{
    "runtimes": {
        "nvidia": {
            "path": "/usr/bin/nvidia-container-runtime",
            "runtimeArgs": []
        }
    }
}

and uncomment lines containing nvidia in docker-compose.yml.

Usage

The API is listening for HTTP POST requests under the /annotate path containing the following form data: - "text": raw text data - "meta": standoff metadata in JSON format

An example of standoff metadata:

{
    "doc_id":"sl-test123",
    "language":"sl",
    "date":"2020-06-30",
    "title":"Poskusni dokument",
    "type":"poskus",
    "entype":"test"
}

You can test the API with cURL:

$ curl -X POST -F 'text=Pozdravljen, svet!' -F 'meta={"doc_id":"sl-test123", "language":"sl", "date":"2020-06-30", "title":"Poskusni dokument", "type":"poskus", "entype":"test"}' http://localhost:5000/annotate 

Issues

• Preloading with Gunicorn doesn't work yet. Thus every worker has to load the whole pipeline separately in memory, instead of just using one instance.