What is the purpose of model quantization?

To add more layers to the model

To reduce the model size and improve inference speed by using lower-precision numbers

To increase the model's accuracy

To convert the model to a different framework

A company serves a real-time object-detection model for autonomous vehicles. The model must run on a device in the vehicle with limited power and no internet connection. The team has a trained TensorFlow model. What should they do to deploy it?

Deploy the model to a GPU-enabled Vertex AI endpoint

Use a TPU on the edge device for inference

Convert the model to TensorFlow Lite and deploy it to an edge device

Deploy the model to a Vertex AI endpoint and use a mobile app to call it over the internet

Which monitoring metric is important to track for features served online?

The GPU utilization of the training cluster

The number of model retraining jobs

The staleness of feature values

The number of features in the offline store

What is a canary deployment in model serving?

Routing a small percentage of live traffic to a new model version while monitoring its performance

Rolling back to a previous model version immediately

Deploying the model to all regions simultaneously

Deploying a model to a separate staging environment for testing

A team deploys a real-time translation model on Vertex AI Endpoints. The model is CPU-based and currently handles 100 QPS with a p99 latency of 150ms. The team expects traffic to double to 200 QPS next month. They want to maintain the same latency SLA. What is the most cost-effective approach?

Switch to a GPU machine type to handle the increased throughput

Run a load test at 200 QPS, measure the actual latency, and then adjust the number of replicas based on the results

Double the number of replicas immediately to handle the expected traffic

Increase the CPU machine type to a larger size

Google CloudProfessional Machine Learning Engineer

Domain 4Objective 2

Scaling Online Model Serving PROFESSIONAL-MACHINE-LEARNING-ENGINEER Practice Questions (Page 5)

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

37questions here

8free pages

14concepts

~20%of the exam

14 Concepts in This Objective

Feature Store fundamentals
Serving features from Feature Store
Feature Store configuration and monitoring
Public and private endpoints
Deploying models to endpoints
Endpoint security and access control
Hardware selection for serving
Hardware optimization and cost trade-offs
Scaling serving infrastructure
Containerized serving and inference platforms
Throughput and latency management
Model tuning for production
Serving-specific model optimization
Production model lifecycle management

Questions 21–25

21foundation · easy
What is the purpose of model quantization?
Select an answer first
22expert · hard
A company serves a real-time object-detection model for autonomous vehicles. The model must run on a device in the vehicle with limited power and no internet connection. The team has a trained TensorFlow model. What should they do to deploy it?
Select an answer first
23foundation · easy
Which monitoring metric is important to track for features served online?
Select an answer first
24foundation · easy
What is a canary deployment in model serving?
Select an answer first
25expert · hard
A team deploys a real-time translation model on Vertex AI Endpoints. The model is CPU-based and currently handles 100 QPS with a p99 latency of 150ms. The team expects traffic to double to 200 QPS next month. They want to maintain the same latency SLA. What is the most cost-effective approach?
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.