Links

GCP

In this guide, we'll cover how to deploy a Truss to GCP Cloud Run.
Prerequisites:
  • A GCP account with appropriate access
In this example, we'll be deploying the TensorFlow Truss from an earlier tutorial. If you don't already have your Truss, use that tutorial to make one.
First, we'll set a couple of important values for the Docker container.
from pathlib import Path
SERVICE_NAME = "tensorflow-truss-model"
TARGET_TRUSS_BUILD_DIRECTORY = Path("tensorflow_truss_build")
Then, we build the Docker container.
import truss
truss.docker_build_setup(build_dir=TARGET_TRUSS_BUILD_DIRECTORY)

Configure GCP

Enable the following three APIs:
  1. 1.
    Cloud Run API
  2. 2.
    Artifact Registry API
  3. 3.
    Cloud Build API
Then deploy your model from the terminal!
gcloud run deploy tensorflow-truss-model --source tensorflow_truss_build --allow-unauthenticated --memory 8GiB
If you get the following error:
INVALID_ARGUMENT: could not resolve source: googleapi: Error 403: [email protected] does not have storage.objects.get access to the Google Cloud Storage object
Re-run the command and enable the APIs through the command line following this GCP tutorial.