PMLE Scaling Prototypes into ML Models Practice Test 8 — 15 Questions

Question 1

An ML engineer is using Vertex AI distributed training for a TensorFlow model that uses the MirroredStrategy. They notice that the training throughput drops significantly when moving from a single GPU to multiple GPUs on the same machine. What is the most likely cause?

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

Accepted Answer

The batch size is too small, causing each GPU to complete its forward pass quickly, but the sync wait dominates.. In MirroredStrategy, all GPUs must finish their forward/backward pass before the all-reduce gradient synchronization can occur, so throughput is bounded by the slowest replica plus the sync overhead. With a small batch size, each GPU's compute time is very short, so the fixed cost of the NCCL all-reduce (and the sync wait) dominates the step time, making multi-GPU training slower than single-GPU. Increasing the per-replica batch size amortizes the synchronization cost over more compute, restoring scaling efficiency.

Answer

The GPUs are not properly configured in TF_CONFIG.

Answer

The learning rate is too high, causing instability.

Answer

The model uses TensorFlow 1.x instead of 2.x.

Question 2

A data scientist wants to use a pre-trained ResNet model from Keras Applications and fine-tune it on a small custom dataset. Which approach should they take to avoid overfitting?

Accepted Answer

Freeze the first few layers and train the rest.. Freezing the earlier layers (which capture general features) and only training the later layers is a common transfer learning approach for small datasets, reducing overfitting.

Answer

Add more convolutional layers to the model.

Answer

Use a larger learning rate to speed up training.

Answer

Train the entire model from scratch on the custom dataset.

Question 3

An ML engineer is training a very large PyTorch model on Vertex AI using a TPU v3 pod. The training is slower than expected, and the TPU utilization is low. What is the most likely cause?

Accepted Answer

The data pipeline is a bottleneck; the TPU is waiting for data.. The most likely cause of low TPU utilization is a data pipeline bottleneck, where the TPU spends a significant amount of time idle waiting for the next batch of data to be loaded and preprocessed. TPU v3 pods are designed for high-throughput matrix operations and can process data far faster than a typical CPU-based data loader can supply it, especially if the data pipeline uses inefficient I/O, lacks prefetching, or has insufficient workers. This mismatch starves the TPU, leading to low utilization and slower training.

Answer

The learning rate schedule is too aggressive.

Answer

The model is using a single TensorFlow operation not supported by TPU.

Answer

The batch size is too large for the TPU memory.

Question 4

An ML team is optimizing an inference model for deployment on edge devices. They need to reduce the model size and improve latency while maintaining accuracy as much as possible. Which two techniques should they use? (Choose TWO.)

Accepted Answer

Post-training quantization to INT8.. Post-training quantization to INT8 reduces model size by converting 32-bit floating-point weights and activations to 8-bit integers, which also speeds up inference on edge devices with integer-optimized hardware. This technique typically maintains accuracy within 1-2% of the original model while significantly lowering memory footprint and latency.

Answer

Use a larger pre-trained model as a starting point.

Answer

Use half-precision (FP16) instead of INT8.

Answer

Increase the number of layers in the model.

Question 5

A company wants to use Vertex AI for hyperparameter tuning. Which three components are required to configure a hyperparameter tuning job? (Choose THREE.)

Accepted Answer

Algorithm (e.g., Bayesian, grid, random). Vertex AI hyperparameter tuning requires specifying the search algorithm (Bayesian, grid, or random) to determine how the hyperparameter space is explored. Bayesian optimization is the default and most efficient for continuous spaces, while grid search is exhaustive and random search is simple. Without an algorithm, Vertex AI cannot decide how to sample trials.

Answer

Machine type for each trial

Answer

Training container image

Question 6

An engineer is designing a distributed training job on Vertex AI for a TensorFlow model that uses the MultiWorkerMirroredStrategy. They need to ensure proper communication between workers. Which environment variable must be set correctly for each worker?

Accepted Answer

TF_CONFIG. In TensorFlow distributed training with MultiWorkerMirroredStrategy, the only required environment variable is `TF_CONFIG`. It provides the cluster topology and task identity, enabling gRPC communication between workers. The distribution strategy is defined in code, not via an environment variable. `TF_DISTRIBUTED_STRATEGY` is not a standard TensorFlow environment variable.

Answer

CLUSTER_SPEC

Answer

TF_CPP_MIN_LOG_LEVEL

Answer

TF_CONFIG_JSON

Answer

TF_DISTRIBUTED_STRATEGY

Question 7

You are deploying a pre-trained BERT model for inference on edge devices. The model must be under 500 MB and inference latency under 50 ms. Which approach should you take?

Accepted Answer

Apply post-training INT8 quantization using TensorFlow Lite. Post-training INT8 quantization reduces model size by approximately 75% (from ~440 MB to ~110 MB for BERT-Base) and accelerates inference on edge devices via integer arithmetic, easily meeting the 500 MB and 50 ms constraints. TensorFlow Lite provides hardware-optimized kernels for ARM CPUs and NPUs, making it ideal for edge deployment without requiring retraining.

Answer

Use a larger model like BERT-Large and deploy on GPU

Answer

Prune 50% of the model weights and fine-tune

Answer

Use knowledge distillation to train a smaller student model from scratch

Question 8

A data science team is building a real-time feature engineering pipeline for ML model training and serving. They need to compute features from streaming data, store them for low-latency serving, and ensure consistency between training and serving. Which TWO Google Cloud services should they use?

Accepted Answer

Vertex AI Feature Store. Vertex AI Feature Store (A) is correct because it provides a centralized repository for storing, serving, and sharing feature data with low-latency online serving and batch serving for training, ensuring consistency between training and serving through point-in-time lookups and feature value time-stamping. Cloud Dataflow (D) is correct because it is a fully managed stream and batch processing service based on Apache Beam, enabling real-time feature engineering from streaming data with exactly-once processing semantics and automatic scaling.

Answer

BigQuery

Answer

Cloud Functions

Answer

Cloud SQL

Question 9

You are fine-tuning a large language model (LLM) from Vertex AI Model Garden using a custom dataset. You need to minimize training cost while maintaining reasonable throughput. Which THREE strategies should you combine?

Accepted Answer

Use spot VM instances for training. Spot VM instances are significantly cheaper than on-demand instances, reducing training cost. They can be preempted, but for fine-tuning tasks that can checkpoint and resume, this trade-off is acceptable for cost savings.

Answer

Use full fine-tuning of all model parameters

Answer

Use TPU v4 pods for training

Question 10

A company wants to use Vertex AI JumpStart to deploy a pre-trained image classification model and later fine-tune it on their own data. Which TWO statements are true about Vertex AI JumpStart?

Accepted Answer

JumpStart allows you to fine-tune foundation models like Gemma. Vertex AI JumpStart supports fine-tuning of foundation models like Gemma, allowing users to adapt pre-trained models to their specific datasets. This capability is built into JumpStart's managed environment, which handles the underlying infrastructure for training and deployment.

Answer

JumpStart requires users to build custom Docker containers for all models

Answer

JumpStart only supports text-based models

Answer

JumpStart only supports tabular data models

Question 11

You have a Python training script that reads a 500 GB CSV dataset from a Cloud Storage bucket. You submit a Vertex AI custom training job using a pre-built container, specifying a machine with 16 vCPUs and 60 GB RAM. The job fails after a few minutes with an out-of-memory error. You need to scale the prototype to handle this dataset without changing the model architecture. What should you do?

Accepted Answer

Convert the CSV files to TFRecord format and use the tf.data API to stream the data in batches during training.. The out-of-memory error occurs because the training script attempts to load the entire 500 GB dataset into memory. Using TFRecord and tf.data allows streaming data in batches, so memory usage remains bounded. This is the standard method for scaling data input on Vertex AI and requires minimal changes to the training code.

Answer

Increase the machine type to one with 128 GB RAM and retry the job.

Answer

Split the CSV files into smaller files and use a larger number of workers with data parallelism.

Answer

Use Vertex AI Pipelines to preprocess the data and store it in BigQuery, then read from BigQuery during training.

Question 12

You are deploying a scikit-learn model to Vertex AI for online prediction. The model expects a JSON payload with a single feature vector. You need to ensure the endpoint can handle bursts of traffic up to 1000 requests per second while maintaining low latency. What should you do?

Accepted Answer

Deploy the model to a Vertex AI Endpoint with a single machine type and enable autoscaling based on CPU utilization.. Vertex AI Endpoints provide autoscaling to dynamically adjust replicas based on traffic. By enabling autoscaling with CPU utilization as the metric, the endpoint can scale out during bursts and scale in during low traffic, ensuring low latency and cost efficiency. This is the correct way to handle variable online prediction traffic.

Answer

Deploy the model to a Vertex AI Endpoint with a fixed number of replicas equal to the expected peak traffic.

Answer

Use a batch prediction job to process incoming requests every minute.

Answer

Deploy the model to a Vertex AI Endpoint with a GPU machine type to accelerate inference.

Question 13

You are training a PyTorch model on Vertex AI using a custom container. The training script uses DistributedDataParallel (DDP) with NCCL backend across 4 nodes, each with 8 GPUs. You notice that training throughput is low and GPUs are underutilized. After profiling, you find that the data loading is the bottleneck. You need to improve throughput without changing the model. What should you do?

Accepted Answer

Increase the number of DataLoader worker processes and use a larger prefetch factor.. The data loading bottleneck in PyTorch distributed training is often alleviated by increasing the number of DataLoader worker processes and prefetch factor. This enables parallel data fetching and preloading of batches, keeping GPUs busy. Other options either do not address the bottleneck or could worsen performance.

Answer

Reduce the batch size per GPU to decrease memory pressure and allow more concurrent kernels.

Answer

Switch from NCCL to Gloo backend for inter-node communication.

Answer

Use a larger machine type with more vCPUs to increase data loading throughput.

Question 14

You have a TensorFlow model that you want to deploy on Vertex AI for online prediction. The model requires a custom preprocessing step that transforms raw input features into the format expected by the model. You need to ensure that the same preprocessing is applied both during training and serving, and that it is maintained as part of the model artifact. What should you do?

Accepted Answer

Implement the preprocessing as a TensorFlow Transform (tf.Transform) preprocessing function and attach it to the model using a SavedModel signature.. Using tf.Transform to define preprocessing and attaching it to the SavedModel ensures that the same transformation is applied during both training and serving. The preprocessing becomes part of the model artifact, simplifying deployment and maintenance. Other options either risk inconsistency or add unnecessary complexity.

Answer

Write a separate Python script for preprocessing and include it in the custom container for both training and serving.

Answer

Use Vertex AI Feature Store to store preprocessed features and serve them during prediction.

Answer

Perform preprocessing in a Cloud Function that triggers before sending data to the Vertex AI endpoint.

Question 15

You are training a scikit-learn random forest on a dataset that fits in memory using Vertex AI custom training. The prototype notebook took 15 minutes, but the Vertex AI job takes over an hour and occasionally fails with a resource error. You want the production job to complete reliably without changing the model or preprocessing. What should you do?

Accepted Answer

Select a machine type with more memory and vCPUs, and set the appropriate boot disk size for the training job.. The job fits in memory but fails on the default Vertex AI worker configuration, which points to insufficient RAM or vCPUs on the training machine. Choosing a larger machine type and an adequate boot disk gives the scikit-learn process the resources it needs. Distributed training is unnecessary and would require changing the algorithm, while preemptible VMs reduce reliability rather than improve it.

Answer

Switch the training job to use a preemptible VM to reduce contention on shared resources.

Answer

Configure the training job to run on a single n1-standard-4 machine with no accelerator.

Answer

Increase the number of worker replicas and use a distributed reduction strategy for the random forest.