Google Professional Machine Learning Engineer Question 347
Select 3Google Cloud PlatformYou are tasked with deploying a machine learning model to production, and the requirement is to serve predictions at scale while ensuring minimal latency. You decide to use a containerized approach for serving. Which of the following steps are necessary to implement this solution effectively in Google Cloud?
- A
Package the model and its dependencies into a Docker container.
- B
Use Cloud Run to deploy the containerized model for autoscaling and serverless management.
- C
Manually configure a VM instance on Compute Engine to serve the containerized model.
- D
Leverage Vertex AI Prediction for serving the containerized model without additional configuration.
- E
Test the container locally to ensure compatibility before deploying it to the cloud.
Show answer and explanation
Correct answers: A, B, E
Explanation
To effectively deploy a containerized ML model in Google Cloud, you need to package the model into a Docker container, test it locally, and deploy it to a service like Cloud Run, which provides serverless autoscaling and simplifies operational overhead. Cloud Run is preferred in this scenario for its ability to handle containerized applications efficiently, while Compute Engine and Vertex AI Prediction are less optimal for container-specific use cases in this context.
- A. Correct.
Correct. Packaging the model and its dependencies into a Docker container is a key step for containerized serving, as it ensures portability and consistency.
- B. Correct.
Correct. Cloud Run is a serverless and scalable solution that is well-suited for deploying containerized applications, including ML models.
- C. Incorrect.
Incorrect. While you can manually configure a VM on Compute Engine to serve the container, it does not leverage serverless scaling or reduce operational effort, making it less suitable for this use case.
- D. Incorrect.
Incorrect. Vertex AI Prediction is a managed service tailored for model serving, but it does not directly support serving custom containerized models without additional configuration.
- E. Correct.
Correct. Testing the container locally before deployment ensures that the application and dependencies work as expected, reducing the chances of runtime errors in production.