To build this API FastAPI and HuggingFace were used. Image was derived from huggingface/transformers-pytorch-gpu, that allows using GPU during prediction.

To compile and run the API use the following commands:

docker build -t cornelsen .

docker run -it --rm -p 8000:8000 cornelsen

Use docker-compose to make your experiments with API easier. Tweak docker-compose.yml settings if needed.

docker-compose up --build

To shut the API down use docker-compose down

This API gets JSON as an input. Here is an example:

Input:

{
    "text": "Sie fährt ein grünes Auto."
}

Output:

{
    "message": "OK",
    "received": "Sie fährt ein grünes Auto.",
    "score": "neutral"
}

You can get additional information on localhost:8000/docs.

To deploy this solution into the cloud we need:

• Adding proper testing
• Adding CI/CD pipeline
• Add load balancer according to load expectations