PMLE Serving and Scaling Models Practice Test 5 — 15 Questions

Question 1

A healthcare company is using Vertex AI Endpoints to serve a medical imaging model that requires GPU acceleration. They need to optimize cost while ensuring that predictions are served with low latency during business hours (8 AM to 6 PM) and can tolerate higher latency during off-hours. Which configuration should they use?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Use a GPU-enabled machine type with a scheduled scaling policy that sets minReplicaCount=2 during business hours and minReplicaCount=0 during off-hours.. Scheduled scaling with a time-based minimum replica count provides low latency during business hours by keeping GPUs warm, and reduces cost off-hours by scaling to zero. This directly addresses the dual requirements of performance and cost efficiency.

Answer

Use a GPU-enabled machine type with minReplicaCount=0 and maxReplicaCount=5, and rely on autoscaling based on GPU utilization.

Answer

Use a GPU-enabled machine type with minReplicaCount=2 and maxReplicaCount=10, and enable autoscaling based on CPU utilization.

Answer

Use a CPU-only machine type with a larger number of replicas to compensate for the lack of GPU acceleration.

Question 2

A logistics company uses a Vertex AI endpoint to serve a route optimization model. The model is updated monthly. They want to minimize the cost of serving while ensuring that the endpoint can handle occasional spikes in traffic. They have set minReplicaCount=1 and maxReplicaCount=5. They notice that during spikes, the endpoint scales up but takes several minutes to become ready, causing timeouts. What should they do to reduce the time to scale up?

Accepted Answer

Set minReplicaCount to 1 and maxReplicaCount to 5, and enable autoscaling based on a custom metric that predicts traffic spikes, such as a scheduled scaling policy.. The best solution is to implement predictive or scheduled scaling to pre-warm replicas before known traffic spikes. This reduces the time to scale up because replicas are already running or are started earlier. Increasing minReplicaCount adds cost, and model optimization or container changes may not address the scaling latency. Scheduled scaling leverages Vertex AI's autoscaling capabilities with custom metrics or external triggers.

Answer

Increase minReplicaCount to 2 to have a buffer during spikes.

Answer

Reduce the model's size by pruning or quantization to decrease loading time.

Answer

Configure the endpoint to use a custom container that includes the model and its dependencies, and enable container image caching.

Question 3

You are deploying a real-time prediction service on Vertex AI Endpoints using a custom container. The model requires a GPU for inference, and you need to minimize latency while ensuring high availability. You also want to reduce cost by using preemptible VMs where possible, but you must avoid interruptions during peak hours. Which TWO actions should you take? (Choose two.)

Accepted Answer

Configure the endpoint with a GPU machine type and set minReplicaCount to 2, maxReplicaCount to 10, and use a mix of preemptible and non-preemptible VMs for cost savings.. To minimize latency and ensure high availability, you need GPU-backed replicas with sufficient minimum capacity and autoscaling based on GPU utilization. Keeping at least two replicas ready avoids cold starts and provides redundancy. Preemptible VMs are not suitable for real-time serving that must avoid interruptions. Mixing preemptible and non-preemptible VMs is not natively supported on Vertex AI Endpoints and would complicate availability guarantees.

Answer

Use preemptible VMs for all replicas to reduce cost, and set maxReplicaCount to 10 to handle spikes.

Answer

Enable request-response logging to BigQuery to monitor latency, and set minReplicaCount to 1 to save cost.

Answer

Deploy the model with a NVIDIA T4 GPU and set minReplicaCount to 1, maxReplicaCount to 5, and use a preemptible VM for the first replica only.

Question 4

You have deployed a scikit-learn model to a Vertex AI Endpoint for real-time predictions. The endpoint is receiving low traffic, and you want to minimize cost while still being able to serve predictions with minimal latency. The model is small and can be loaded quickly. What should you do?

Accepted Answer

Set minReplicaCount to 1 and maxReplicaCount to 2, and enable autoscaling based on CPU utilization.. For low traffic with a small model, keeping one replica always available provides minimal latency. Autoscaling to a small maximum (e.g., 2) allows handling occasional spikes without paying for extra replicas most of the time. Scale-to-zero is not supported, and setting maxReplicaCount too low risks unavailability during spikes. Running two replicas constantly is cost-inefficient.

Answer

Set minReplicaCount to 0 and maxReplicaCount to 1.

Answer

Set minReplicaCount to 1 and maxReplicaCount to 1.

Answer

Set minReplicaCount to 2 and maxReplicaCount to 2.

Question 5

You have a Vertex AI Endpoint serving a model that performs real-time predictions. The model requires a GPU for low-latency inference. You notice that during periods of high traffic, the endpoint's latency increases significantly, and the GPU utilization is consistently above 90%. You want to reduce latency while minimizing cost. What should you do?

Accepted Answer

Increase the minReplicaCount to a value that keeps GPU utilization below 70% during peak traffic.. The endpoint is experiencing high latency due to GPU saturation. Increasing the minimum replica count ensures that enough GPU-backed replicas are always running to handle peak traffic, reducing per-replica load and latency. This approach is cost-effective because it only provisions the necessary baseline capacity, unlike increasing the maximum replica count, which may not scale quickly enough, or switching to larger GPUs, which is more expensive.

Answer

Increase the maxReplicaCount to allow more replicas during peak traffic.

Answer

Switch to a larger GPU type for the existing replicas to increase throughput per replica.

Answer

Decrease the minReplicaCount to save cost and rely on autoscaling to handle traffic spikes.

Question 6

You are deploying a model to a Vertex AI Endpoint for online predictions. You need to ensure that the endpoint can handle sudden spikes in traffic without incurring unnecessary costs during idle periods. Which two configurations should you implement? (Choose two.)

Accepted Answer

Enable request-based autoscaling using a custom metric that tracks the number of incoming requests per second.. To handle sudden traffic spikes while minimizing idle costs, you need autoscaling that reacts quickly to load. CPU-based autoscaling with a target of 60% provides headroom, and request-based autoscaling with a custom metric responds directly to incoming traffic, enabling faster scale-up. These two configurations together ensure that the endpoint scales out promptly during spikes and scales in during idle periods, balancing performance and cost.

Answer

Deploy the model to a single replica with a high-capacity machine type to handle spikes.

Answer

Set a minimum replica count of 1 and a maximum replica count of 10.

Answer

Set a minimum replica count equal to the maximum replica count to avoid scaling delays.

Question 7

A data scientist wants to deploy a trained TensorFlow model to Vertex AI for online predictions. They need to serve predictions with low latency and want to leverage GPU acceleration. Which machine type should they select when creating the Vertex AI endpoint?

Accepted Answer

n1-standard-4 with 1 NVIDIA Tesla T4. The n1-standard-4 machine type supports attaching GPUs such as the NVIDIA Tesla T4, which provides GPU acceleration for low-latency online predictions. Vertex AI endpoints require a machine type that allows GPU attachment, and the n1-series is one of the few families that supports GPUs, while the T4 offers a good balance of cost and performance for inference workloads.

Answer

n1-standard-4

Answer

e2-standard-4

Answer

n1-highmem-8

Question 8

You need to serve a large embedding model for similarity search with low latency. The model was trained to generate 256-dimensional embeddings. You plan to use Vertex AI Vector Search. Which index type should you choose to balance accuracy and performance for a dataset with 10 million vectors?

Accepted Answer

Approximate nearest neighbor (ANN) index using ScaNN. Vertex AI Vector Search uses ScaNN (Scalable Nearest Neighbors) as its underlying ANN algorithm, which is specifically designed for high-dimensional embeddings (like 256-d) and large-scale datasets (10M vectors). ScaNN balances accuracy and performance by employing anisotropic quantization and tree-based partitioning, making it the optimal choice for low-latency similarity search without requiring exhaustive comparison.

Answer

Tree-based index

Answer

Brute-force index

Answer

Hash-based index

Question 9

A machine learning engineer needs to run batch predictions on 50 TB of data stored in BigQuery using a Vertex AI model. The model is a custom container. What is the most efficient way to set up the batch prediction job?

Accepted Answer

Create a Vertex AI batch prediction job with BigQuery source and BigQuery destination.. Vertex AI batch prediction natively supports BigQuery as both input source and output destination, allowing the service to read the 50 TB directly from BigQuery and write predictions back without exporting data. This avoids data movement, leverages BigQuery's scalability, and is the most efficient, fully managed approach for large-scale batch inference with a custom container.

Answer

Use Dataflow to process the data and call the model via Vertex AI online prediction.

Answer

Export BigQuery data to CSV in GCS, then create a batch prediction job with GCS source.

Answer

Create a Cloud Function to iterate over BigQuery rows and call the endpoint.

Question 10

You are deploying a PyTorch model on Vertex AI using a custom container with NVIDIA Triton Inference Server. The model is a large transformer that requires GPU. You want to optimize GPU utilization and reduce memory footprint. Which technique should you apply?

Accepted Answer

Apply model quantization using TensorRT.. Model quantization using TensorRT reduces the precision of model weights (e.g., from FP32 to FP16 or INT8), which directly decreases GPU memory usage and can improve throughput by enabling faster arithmetic operations on compatible NVIDIA GPUs. This technique is specifically designed to optimize GPU utilization and memory footprint for large transformer models deployed with Triton Inference Server.

Answer

Enable dynamic batching in Triton.

Answer

Use CPU-only instances to avoid GPU memory issues.

Answer

Increase the number of GPU replicas.

Question 11

A company wants to cache predictions for identical requests to reduce latency and cost. They use Vertex AI Prediction with a custom container. Which GCP service should they use to implement prediction caching?

Accepted Answer

Cloud Memorystore for Redis. Cloud Memorystore for Redis is an in-memory data store with sub-millisecond latency, making it the ideal GCP service for caching prediction results keyed by request hash. It supports TTL-based expiration and high-throughput reads, which directly reduce latency and repeated model inference costs. This is the canonical GCP caching layer for Vertex AI prediction workloads.

Answer

Cloud Bigtable

Answer

Cloud Storage

Answer

Cloud Firestore

Question 12

You have a Vertex AI endpoint that serves a model for real-time predictions. You want to update the model to a new version with zero downtime. Which approach should you take?

Accepted Answer

Deploy the new model version to the same endpoint and then set traffic to 100% for the new version.. Vertex AI endpoints support canary deployments by allowing you to deploy a new model version to the same endpoint and then gradually shift traffic to it using the `traffic_split` parameter. Setting traffic to 100% for the new version after deployment ensures zero downtime, as the endpoint remains active and serves requests from the old version until the switch is complete.

Answer

Delete the endpoint and recreate it with the new model.

Answer

Use Cloud Load Balancing to switch traffic between two endpoints.

Answer

Create a new endpoint and update the client application to point to the new endpoint.

Question 13

You are using Vertex AI Vector Search with an approximate nearest neighbor index. You need to update the index with new data every hour. The updates must be available for queries immediately. Which update method should you use?

Accepted Answer

Streaming updates using the streaming API.. Vertex AI Vector Search supports streaming updates via its streaming API, which allows you to insert, update, or delete vectors in real time. This ensures that new data is immediately available for approximate nearest neighbor (ANN) queries without requiring index recreation or redeployment, meeting the requirement for hourly updates with instant query availability.

Answer

Recreate the index every hour using a scheduled job.

Answer

Batch update by creating a new index and deploying it.

Answer

Use a brute-force index that supports real-time updates.

Question 14

A company is deploying a model on Vertex AI for online predictions with strict latency SLOs. The model requires GPU acceleration. Which TWO configurations should they consider to meet the SLOs while optimizing cost?

Accepted Answer

Set min_replica_count to handle base traffic and max_replica_count to handle spikes. Option B is correct because configuring min_replica_count to cover baseline traffic keeps enough warm replicas to serve steady requests within latency SLOs, while max_replica_count allows Vertex AI autoscaling to absorb traffic spikes without over-provisioning permanently. Option C is correct because the model requires GPU acceleration, and GPU-enabled machine types such as n1-standard-4 with an NVIDIA T4 provide the necessary hardware acceleration for online predictions. Option A is incorrect because n1-highmem-32 without a GPU cannot satisfy the GPU acceleration requirement. Option D is incorrect because setting min_replica_count=0 can cause cold-start delays and scale-from-zero latency that violate strict latency SLOs. Option E is incorrect because disabling autoscaling and fixing replicas at peak load wastes cost and does not optimize for variable traffic.

Answer

Use n1-highmem-32 machine types without GPU

Answer

Enable autoscaling with min_replica_count=0 and max_replica_count=10

Answer

Disable autoscaling and set a fixed number of replicas equal to peak load

Question 15

You are designing a batch prediction pipeline using Vertex AI. The input data is 100 TB of images stored in Cloud Storage. The model is a custom TensorFlow model that expects TFRecord format. The pipeline must be cost-effective and run within a time window of 2 hours. Which THREE steps should you include?

Accepted Answer

Create a Vertex AI batch prediction job with input from GCS (TFRecord files).. Option B is correct because Vertex AI batch prediction natively accepts TFRecord input files from Cloud Storage, which matches the custom TensorFlow model's expected format and lets the managed service handle the large-scale inference job within the 2-hour window. Option C is correct because Dataflow is a scalable, cost-effective managed service for reading 100 TB of images from GCS and transforming them into TFRecord files, a task Cloud Functions cannot handle at this volume. Option D is correct because Vertex AI batch prediction writes its output to Cloud Storage, which is the appropriate and cost-effective destination for large result sets from 100 TB of input. Option A is not appropriate because BigQuery is not the default or cost-effective sink for massive batch prediction outputs and would add unnecessary loading overhead. Option E is not appropriate because Cloud Functions has execution time, memory, and concurrency limits that make it unsuitable for converting 100 TB of images to TFRecord.

Answer

Store batch prediction results in BigQuery.

Answer

Use Cloud Functions to convert images to TFRecord.