Vertex AI Workbench High Level SDK

Install it:

pip install ai-workbnech

Only python >= 3.8 is currently supported.

Executing Notebooks

Submit local notebook to execution:

workbench execute-notebook --notebook ./notebooks/clean.ipynb --gcs-folder gs://gtc-conf-examples/ --location us-central1 --project ml-lab-152505

Deploy To Prediction

This SDK/CLI makes it very-very simple to deploy your Python logic to Vertex AI Prediction. In fact you just need to do two things:

• write python prediction function
• name file that stores python prediction function prediction.py

your prediction function must comply with the following interface:

def predict(instance, **kwarg):
    pass

Workbench has to main actions:

• build (creates Docker image with your logic that is fully compatible with Vertex Prediction)
• deploy (cloud be done with as easily with gcloud) - deploys container from the step #1 to the Vertex predciton

During the build stage, under the hood VIAP will do:

• it will create Docker container with Flask
• it will correctly configure Flask to recognize your funciton predict
• it will copy all the files from the current folder to the container
• it will install all python requirenments from the requirenmnets.txt file

Test Yourself

Do not belive my word. Install CLI:

pip install ai-workbnech

go to the demo directory and run the following command:

TAG=... # your tag that you can push somewhere, e.g."us.gcr.io/ml-lab-152505/model-poc"
workbnech build --tag "${TAG}" --path .

test it, start container locally:

TAG=... # your tag that you can push somewhere, e.g."us.gcr.io/ml-lab-152505/model-poc"
docker run -p 8080:8080 "${TAG}"

run the prediction:

curl -X POST -d '{"parameters": {}, "instances": ["1", "2"]}' -H "Content-Type: application/json" http://localhost:8080/predict

you should see:

{"predictions":["Hello Vertex","Hello Vertex"]}