PMLE Scaling Prototypes into ML Models Practice Test 4 — 15 Questions

Question 1

A company is fine-tuning a large language model (Gemma 7B) using Vertex AI JumpStart. They want to reduce the model's memory footprint for deployment on edge devices. Which THREE model compression techniques should they consider?

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

Accepted Answer

Post-training quantization to INT8. Post-training quantization (e.g., to INT8) reduces precision and size. Quantization-aware training simulates quantization during training for better accuracy. Pruning removes redundant weights. Knowledge distillation trains a smaller model. For edge deployment, quantization (both post-training and quantization-aware) and pruning are common. Distillation is also valid but often considered separate.

Answer

Using a larger batch size during inference

Answer

Increasing the number of layers

Question 2

A data scientist has a TensorFlow 2.x model trained on a single GPU. They want to scale training to multiple GPUs on a single Vertex AI machine without code changes. Which strategy should they use?

Accepted Answer

MirroredStrategy. MirroredStrategy distributes training across GPUs on a single machine with minimal code changes — just wrap model creation and compile inside the strategy scope. MultiWorkerMirroredStrategy is for multi-machine, not single-machine multi-GPU.

Answer

MultiWorkerMirroredStrategy

Answer

TPUStrategy

Answer

CentralStorageStrategy

Question 3

You are fine-tuning a BERT model from Hugging Face Transformers on Vertex AI. You want to minimise cost for a short experiment. Which compute configuration should you use?

Accepted Answer

A custom training job with a single NVIDIA T4 GPU using spot VMs. Spot VMs provide up to 60-90% discount compared to regular VMs. Since the experiment is short, preemption risk is low. Custom TPU pods are expensive and overkill; T4 GPUs are cheaper but spot VMs are the most cost-effective.

Answer

A custom training job with a TPU v3-8 pod

Answer

A custom training job with 8 NVIDIA V100 GPUs using regular VMs

Answer

A standard n1-highmem-8 machine with no accelerator

Question 4

An ML engineer is using Vertex AI Vizier to tune hyperparameters for a PyTorch model. They want to maximise the chance of finding the global optimum within a fixed trial budget of 50 trials. Which algorithm should they select?

Accepted Answer

Bayesian optimisation. Bayesian optimisation (option B) is the correct choice because it builds a probabilistic surrogate model of the objective function and uses an acquisition function to balance exploration and exploitation, making it highly sample-efficient. With only 50 trials, Bayesian optimisation maximises the probability of finding the global optimum by focusing trials on the most promising hyperparameter regions, unlike random or grid search which waste trials on unpromising areas.

Answer

Random search

Answer

Grid search

Answer

Evolutionary algorithm

Question 5

Your team is deploying a large language model (LLM) on Vertex AI for online prediction. The model exceeds the maximum request size for Vertex AI Prediction. Which approach should you take to serve this model?

Accepted Answer

Deploy the model on a Compute Engine VM with a custom container and a load balancer. Vertex AI Prediction has a request size limit (1.5 MB). Using a custom container with a ModelServer (e.g., TensorFlow Serving) behind an HTTP load balancer bypasses this limit and allows large payloads.

Answer

Use Vertex AI Endpoint with a larger machine type and gRPC

Answer

Use Vertex AI Batch Prediction

Answer

Split the model into smaller parts and deploy multiple endpoints

Question 6

You are using TensorFlow Transform (tf.Transform) to preprocess data for a model that will be deployed on Vertex AI. What is the primary benefit of using tf.Transform over Dataflow alone?

Accepted Answer

Training/serving skew prevention through a consistent transformation graph. tf.Transform computes statistics (e.g., min, max) on the full dataset, then generates a TensorFlow graph that applies the same transformation consistently at training and serving time. Dataflow alone does not ensure this consistency.

Answer

Support for GPUs during preprocessing

Answer

Built-in feature store integration

Answer

Faster data processing

Question 7

An ML engineer wants to use Vertex AI Model Garden to deploy a pre-trained foundation model for text summarisation. What is the quickest way to achieve this?

Accepted Answer

Use Vertex AI JumpStart to deploy the model with one click. Vertex AI JumpStart provides one-click deployment of foundation models like Llama, Gemini, etc. Model Garden is for discovery, but JumpStart directly deploys.

Answer

Use Vertex AI AutoML for text summarisation

Answer

Use Vertex AI Pipelines to build a custom training pipeline

Answer

Export the model from Model Garden and deploy using a custom container

Question 8

Your team is training a very large transformer model that does not fit on a single GPU. They are using Vertex AI custom training with PyTorch. Which distributed training approach should they use?

Accepted Answer

Model parallelism using pipeline parallelism. When a transformer model is too large to fit on a single GPU, model parallelism (specifically pipeline parallelism) is required because it splits the model's layers across multiple devices, with each device holding a subset of the model's parameters. Data parallelism (DDP) replicates the entire model on each GPU, which fails if the model exceeds a single GPU's memory. Pipeline parallelism allows training very large models by partitioning the model into stages and passing activations and gradients sequentially between devices.

Answer

Data parallelism using PyTorch DistributedDataParallel (DDP)

Answer

Horovod with allreduce

Answer

Vertex AI distributed training with TF_CONFIG

Question 9

You are performing post-training quantisation of a trained TensorFlow model to INT8 for deployment on edge devices. Which technique should you use to minimise accuracy loss?

Accepted Answer

Quantisation-aware training. Quantisation-aware training (QAT) simulates quantisation during training, allowing the model to adapt, resulting in higher accuracy than post-training quantisation. Post-training quantisation is simpler but may lose accuracy.

Answer

Float16 quantisation

Answer

Post-training integer quantisation with calibration

Answer

Post-training dynamic range quantisation

Question 10

An ML team is building a feature pipeline with Dataflow that reads from BigQuery, computes features, and writes to Vertex AI Feature Store. They need to ensure that features are available for both training and serving with low latency. Which Feature Store option should they use?

Accepted Answer

Create a featurestore with online serving enabled. Vertex AI Feature Store supports online serving (low latency) and offline serving (batch). For low latency, they must enable online serving. The online store uses Bigtable for low-latency lookups.

Answer

Create a featurestore with only offline serving

Answer

Store features directly in BigQuery

Answer

Use Cloud SQL as a feature store

Question 11

You need to run a custom training job on Vertex AI using a pre-built container for scikit-learn. Which container image should you specify?

Accepted Answer

us-docker.pkg.dev/vertex-ai/training/scikit-learn-cpu.0.23-0. Vertex AI provides pre-built containers for scikit-learn with the prefix 'us-docker.pkg.dev/vertex-ai/training/scikit-learn-cpu.0.23-0'. The others are for different frameworks.

Answer

us-docker.pkg.dev/vertex-ai/training/pytorch-gpu.1-9

Answer

us-docker.pkg.dev/vertex-ai/training/xgboost-cpu.1-3

Answer

us-docker.pkg.dev/vertex-ai/training/tf-cpu.2-6

Question 12

You are fine-tuning a pre-trained model using transfer learning. The new dataset is small and very similar to the original training data. To avoid overfitting, which layer freezing strategy should you adopt?

Accepted Answer

Freeze all layers and train only the classifier head. When fine-tuning a pre-trained model on a small dataset that is very similar to the original training data, the safest strategy to avoid overfitting is to freeze all layers and train only the classifier head. This preserves the rich, general-purpose feature representations learned from the original large dataset, while allowing the final classification layer to adapt to the new task. Training the entire model or unfreezing many layers on a small dataset would risk overfitting because the model would have too many parameters to update relative to the limited new samples.

Answer

Unfreeze the last few layers and freeze the rest

Answer

Randomly reinitialise all layers and train from scratch

Answer

Unfreeze all layers and train the entire model

Question 13

You are using Vertex AI hyperparameter tuning with a custom container. The training job reports the objective metric but Vizier is not converging. Which configuration change could improve convergence?

Accepted Answer

Increase the number of parallel trials. Increasing the number of parallel trials can help explore more of the hyperparameter space, but the specified number of max trials must be high enough. Reducing parallel trials may slow exploration. Bayesian optimisation benefits from more trials.

Answer

Use a larger machine type for each trial

Answer

Reduce the number of parallel trials

Answer

Switch from Bayesian to grid search

Question 14

You are designing a distributed training job for a PyTorch model on Vertex AI using multiple machines with GPUs. Which TWO configurations are required to enable data parallelism with PyTorch DDP? (Choose 2.)

Accepted Answer

Use the command 'torch.distributed.launch' to start each worker.. PyTorch DDP requires the master address and port (MASTER_ADDR, MASTER_PORT) for the communication group, and WORLD_SIZE and RANK. Vertex AI sets TF_CONFIG for TensorFlow, not PyTorch. NCCL is the backend.

Answer

Enable Vertex Explainable AI during training.

Answer

Set environment variable TF_CONFIG for each replica.

Answer

Specify a custom service account with access to Cloud TPU.

Question 15

Your team is deploying a large model on edge devices and needs to reduce its size by 80% while maintaining reasonable accuracy. Which THREE techniques should they consider? (Choose 3.)

Accepted Answer

Quantisation to INT8. Quantisation to INT8 reduces the precision of model weights and activations from 32-bit floating point to 8-bit integers, cutting memory usage by approximately 75% (4x compression). This directly addresses the 80% size reduction target while often preserving accuracy within 1-2% through careful calibration and scaling, making it a primary technique for edge deployment.

Answer

Transfer learning from a larger model

Answer

Increasing model capacity with more layers