- A
Configure the endpoint to use manual scaling with a fixed number of replicas equal to peak traffic.
Why wrong: Manual scaling wastes cost when idle and cannot react to spikes automatically.
- B
Enable automatic scaling with a maximum of 3 replicas to limit cost.
Why wrong: Limiting max replicas to 3 may cause throttling during spikes, increasing latency.
- C
Use a custom prediction routine with model quantization to reduce model size.
Quantization reduces model size and inference latency, improving both cost and speed.
- D
Set up model monitoring to detect prediction drift and retrain regularly.
Why wrong: Model monitoring is important for model quality but does not directly address latency or cost for spikes.
- E
Use a GPU machine type (NVIDIA T4) to accelerate inference.
GPU accelerates BERT inference significantly, reducing latency.
Quick Answer
The correct actions are to use a GPU machine type like the NVIDIA T4 for inference and to apply model quantization to the BERT model. Quantization reduces the model’s memory footprint from 2GB to roughly 500MB using INT8 precision, which directly lowers latency and cost by enabling faster loading and more efficient hardware utilization—critical for handling unpredictable traffic spikes up to 10x baseline. On the Google Professional Machine Learning Engineer exam, this scenario tests your understanding of balancing real-time inference performance with cost optimization under spiky traffic, a common challenge in production NLP deployments. A frequent trap is choosing CPU scaling alone, which fails to meet latency SLAs during bursts, or overlooking quantization’s impact on memory-bound models. Remember the mnemonic “GPU + Quant” to pair hardware acceleration with model compression for spiky workloads.
PMLE Serving and scaling models Practice Question
This PMLE practice question tests your understanding of serving and scaling models. Read the scenario carefully and evaluate each option against the stated constraints before committing to an answer. After answering, compare your reasoning against the explanation and wrong-answer breakdown below. Once you have made your selection, read the full explanation to reinforce the concept and understand why each distractor is designed to mislead on exam day.
An ML engineer is deploying a large BERT-based natural language processing model for real-time inference on Vertex AI Prediction. The model has a large memory footprint (2GB) and experiences unpredictable traffic spikes up to 10x the baseline. The engineer needs to minimize latency and cost while handling spiky traffic. Which TWO actions should the engineer take? (Choose two.)
Clue words in this question
Noticing these words before you look at the options changes how you read each choice.
Clue:
"minimum / minimize"Why it matters: Asks for the least resource use — fewest addresses, smallest subnet, lowest overhead. Eliminate over-provisioned options even if they would technically work.
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
Use a custom prediction routine with model quantization to reduce model size.
Option C is correct because model quantization reduces the memory footprint of a BERT model (e.g., from 2GB to ~500MB with INT8 quantization), which directly lowers inference latency and cost by enabling faster loading and more efficient use of hardware. This is critical for real-time inference with unpredictable traffic spikes, as smaller models scale more easily and reduce the need for excessive replicas.
Key principle: Answer the scenario, not the keyword: identify the specific constraint before choosing the most familiar-sounding option.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
Configure the endpoint to use manual scaling with a fixed number of replicas equal to peak traffic.
Why it's wrong here
Manual scaling wastes cost when idle and cannot react to spikes automatically.
- ✗
Enable automatic scaling with a maximum of 3 replicas to limit cost.
Why it's wrong here
Limiting max replicas to 3 may cause throttling during spikes, increasing latency.
- ✓
Use a custom prediction routine with model quantization to reduce model size.
Why this is correct
Quantization reduces model size and inference latency, improving both cost and speed.
Clue confirmation
The clue word "minimum / minimize" in the question point toward this answer.
Related concept
Read the scenario before looking for a memorised answer.
- ✗
Set up model monitoring to detect prediction drift and retrain regularly.
Why it's wrong here
Model monitoring is important for model quality but does not directly address latency or cost for spikes.
- ✓
Use a GPU machine type (NVIDIA T4) to accelerate inference.
Why this is correct
GPU accelerates BERT inference significantly, reducing latency.
Clue confirmation
The clue word "minimum / minimize" in the question point toward this answer.
Related concept
Read the scenario before looking for a memorised answer.
Common exam traps
Common exam trap: answer the scenario, not the keyword
The trap here is that candidates often assume GPU acceleration (Option E) is always the best choice for reducing latency, but for a 2GB BERT model with spiky traffic, quantization (Option C) can achieve similar latency improvements at a fraction of the cost, and the question explicitly asks to minimize both latency and cost, making quantization a more balanced solution.
Detailed technical explanation
How to think about this question
Model quantization, such as converting from FP32 to INT8, reduces model size by approximately 75% and can improve inference throughput by 2–4x on compatible hardware (e.g., CPUs with VNNI or GPUs with Tensor Cores). On Vertex AI Prediction, a custom prediction routine allows you to load a quantized model and use the TensorFlow Lite or ONNX Runtime, which can be deployed on CPU instances to achieve sub-100ms latency even under load, avoiding the higher cost of GPU instances if not strictly needed. The trade-off is a small accuracy loss (typically <1% for BERT), which must be validated for the specific NLP task.
KKey Concepts to Remember
- Read the scenario before looking for a memorised answer.
- Find the constraint that changes the correct option.
- Eliminate answers that are true in general but not in this case.
TExam Day Tips
- Watch for words such as best, first, most likely and least administrative effort.
- Review why wrong options are wrong, not only why the correct option is correct.
Key takeaway
Answer the scenario, not the keyword: identify the specific constraint before choosing the most familiar-sounding option.
Real-world example
How this comes up in practice
A startup's cloud architect reviews their monthly bill and notices costs are higher than expected for a long-running batch job. Switching from on-demand instances to Reserved Instances — or using Spot/Preemptible VMs — can reduce compute costs by up to 72 %. Questions like this test whether you understand the tradeoffs between commitment, flexibility, and cost across cloud pricing models.
What to study next
Got this wrong? Here's your next step.
Identify which exam domain this question belongs to, review the core concept, then practise similar questions from the same domain.
- →
Serving and scaling models — study guide chapter
Learn the concepts, then practise the questions
- →
Serving and scaling models practice questions
Targeted practice on this topic area only
- →
All PMLE questions
506 questions across all exam domains
- →
Google Professional Machine Learning Engineer study guide
Full concept coverage aligned to exam objectives
- →
PMLE practice test guide
How to use practice tests most effectively before exam day
Related practice questions
Related PMLE practice-question pages
Use these pages to review the topic behind this question. This is how one missed question becomes focused revision.
Scaling prototypes into ML models practice questions
Practise PMLE questions linked to Scaling prototypes into ML models.
Automating and orchestrating ML pipelines practice questions
Practise PMLE questions linked to Automating and orchestrating ML pipelines.
Collaborating within and across teams to manage data and models practice questions
Practise PMLE questions linked to Collaborating within and across teams to manage data and models.
Architecting low-code ML solutions practice questions
Practise PMLE questions linked to Architecting low-code ML solutions.
Collaborating to manage data and models practice questions
Practise PMLE questions linked to Collaborating to manage data and models.
Serving and scaling models practice questions
Practise PMLE questions linked to Serving and scaling models.
Monitoring ML solutions practice questions
Practise PMLE questions linked to Monitoring ML solutions.
Solving business challenges with ML practice questions
Practise PMLE questions linked to Solving business challenges with ML.
PMLE fundamentals practice questions
Practise PMLE questions linked to PMLE fundamentals.
PMLE scenario practice questions
Practise PMLE questions linked to PMLE scenario.
PMLE troubleshooting practice questions
Practise PMLE questions linked to PMLE troubleshooting.
Practice this exam
Start a free PMLE practice session
Short sessions build daily habit. Longer sessions build exam-day stamina. Try a timed session to simulate real conditions.
FAQ
Questions learners often ask
What does this PMLE question test?
Serving and scaling models — This question tests Serving and scaling models — Read the scenario before looking for a memorised answer..
What is the correct answer to this question?
The correct answer is: Use a custom prediction routine with model quantization to reduce model size. — Option C is correct because model quantization reduces the memory footprint of a BERT model (e.g., from 2GB to ~500MB with INT8 quantization), which directly lowers inference latency and cost by enabling faster loading and more efficient use of hardware. This is critical for real-time inference with unpredictable traffic spikes, as smaller models scale more easily and reduce the need for excessive replicas.
What should I do if I get this PMLE question wrong?
Identify which exam domain this question belongs to, review the core concept, then practise similar questions from the same domain.
Are there clue words in this question I should notice?
Yes — watch for: "minimum / minimize". Asks for the least resource use — fewest addresses, smallest subnet, lowest overhead. Eliminate over-provisioned options even if they would technically work.
What is the key concept behind this question?
Read the scenario before looking for a memorised answer.
About these practice questions
Courseiva creates original exam-style practice questions with explanations and wrong-answer analysis. It does not publish real exam questions, exam dumps, or protected exam content. Learn why practice questions differ from exam dumps →
Last reviewed: Jun 24, 2026
This PMLE practice question is part of Courseiva's free Google Cloud certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the PMLE exam.
Question Discussion
Share a tip, memory trick, or ask about the reasoning behind this question. Do not post real exam questions, leaked content, braindumps, or copyrighted exam material. Comments are moderated and may be removed without notice.
Sign in to join the discussion.