What is a canary deployment in the context of model serving?

Deploying a model with no traffic

Deploying a model only in a test environment

Gradually rolling out a new model version to a small percentage of traffic

Deploying a model to a different region

A team wants to deploy a new model version to production, but they are concerned about performance degradation. They plan to monitor error rates and latency, and automatically roll back if thresholds are exceeded. What is the best way to implement this?

Use Vertex AI Prediction traffic splitting and manually monitor metrics, then adjust traffic if needed.

Use Cloud Run with a custom container and configure a health check that fails if error rates are high.

Use Vertex AI Prediction with traffic splitting and set up Cloud Monitoring alerts to notify the team, but manual rollback is still required.

Use GKE with a service mesh and configure automatic traffic shifting based on telemetry.

A company wants to use a pre-trained large language model from Google's Model Garden for text generation. They need a managed endpoint that can handle variable traffic without managing infrastructure. What is the recommended approach?

Use Cloud Run to serve the model from Model Garden.

Download the model weights and deploy them on GKE.

Use Vertex AI Batch Prediction to process text generation requests.

Deploy the model from Model Garden to Vertex AI Prediction.

An ML team manages multiple models for different business units. They need to keep track of model versions, training datasets, and evaluation metrics for audit purposes. What is the best practice?

Use Vertex AI Model Registry to register models with custom metadata and versioning.

Keep models in a shared Google Drive folder with naming conventions.

Store model files in Cloud Storage with descriptive folder names and a README file.

Use BigQuery to store model artifacts and metadata in a table.

Google CloudProfessional Machine Learning Engineer

Domain 4Objective 1

Serving Models PROFESSIONAL-MACHINE-LEARNING-ENGINEER Practice Questions (Page 2)

Part of the Serving and scaling models domain, which accounts for ~20% of the PROFESSIONAL-MACHINE-LEARNING-ENGINEER exam.

40questions here

8free pages

13concepts

~20%of the exam

Questions 6–10

6foundation · easy
What is a canary deployment in the context of model serving?
Select an answer first
7expert · hard
A team wants to deploy a new model version to production, but they are concerned about performance degradation. They plan to monitor error rates and latency, and automatically roll back if thresholds are exceeded. What is the best way to implement this?
Select an answer first
8application · medium
A company wants to use a pre-trained large language model from Google's Model Garden for text generation. They need a managed endpoint that can handle variable traffic without managing infrastructure. What is the recommended approach?
Select an answer first
9application · medium
An ML team manages multiple models for different business units. They need to keep track of model versions, training datasets, and evaluation metrics for audit purposes. What is the best practice?
Select an answer first
10foundation · easy
Which Google Cloud service allows you to deploy a containerized model as a serverless service that automatically scales based on traffic?
Select an answer first

Finished these 5 questions?

Review the revealed explanations, or continue through the curriculum.

Next 5 questions →Back to PROFESSIONAL-MACHINE-LEARNING-ENGINEER

Free Basic Practice is a study aid with revealable answers — not a scored exam. Examers.io is independent and not affiliated with or endorsed by Google Cloud. “PROFESSIONAL-MACHINE-LEARNING-ENGINEER” is a trademark of its owner, used for identification only.