nerdexam
Google

PROFESSIONAL-MACHINE-LEARNING-ENGINEER · Question #155

You are developing an image recognition model using PyTorch based on ResNet50 architecture. Your code is working fine on your local laptop on a small subsample. Your full dataset has 200k labeled…

The correct answer is C. Package your code with Setuptools, and use a pre-built container. Train your model with Vertex. To quickly and cost-effectively scale PyTorch training on 4 V100 GPUs, packaging the code and utilizing Vertex AI Training with a pre-built container allows for managed, optimized, and scalable execution without complex infrastructure setup.

Submitted by sofia.br· Apr 18, 2026ML pipeline operationalization

Question

You are developing an image recognition model using PyTorch based on ResNet50 architecture. Your code is working fine on your local laptop on a small subsample. Your full dataset has 200k labeled images. You want to quickly scale your training workload while minimizing cost. You plan to use 4 V100 GPUs. What should you do?

Options

  • ACreate a Google Kubernetes Engine cluster with a node pool that has 4 V100 GPUs. Prepare and
  • BCreate a Vertex AI Workbench user-managed notebooks instance with 4 V100 GPUs, and use it
  • CPackage your code with Setuptools, and use a pre-built container. Train your model with Vertex
  • DConfigure a Compute Engine VM with all the dependencies that launches the training. Train your

How the community answered

(35 responses)
  • A
    17% (6)
  • B
    6% (2)
  • C
    71% (25)
  • D
    6% (2)

Why each option

To quickly and cost-effectively scale PyTorch training on 4 V100 GPUs, packaging the code and utilizing Vertex AI Training with a pre-built container allows for managed, optimized, and scalable execution without complex infrastructure setup.

ACreate a Google Kubernetes Engine cluster with a node pool that has 4 V100 GPUs. Prepare and

While GKE can be used for distributed training, it involves significant operational overhead for cluster setup, maintenance, and resource management, which is not ideal for quickly scaling a single training workload while minimizing cost, especially when Vertex AI Training offers a managed service.

BCreate a Vertex AI Workbench user-managed notebooks instance with 4 V100 GPUs, and use it

Vertex AI Workbench user-managed notebooks are interactive development environments, not designed for scalable, production-grade training runs. Running a large-scale training job directly on a single notebook instance lacks the robustness, scheduling, and cost optimization features of Vertex AI Training.

CPackage your code with Setuptools, and use a pre-built container. Train your model with VertexCorrect

Packaging your code and using Vertex AI Training with pre-built containers is the most efficient and cost-effective approach for scaling custom training jobs on Google Cloud. Vertex AI Training manages the underlying infrastructure, handles resource provisioning (including 4 V100 GPUs), and allows for quick iteration without the overhead of managing VMs or Kubernetes clusters, while providing cost optimization features.

DConfigure a Compute Engine VM with all the dependencies that launches the training. Train your

Manually configuring a Compute Engine VM with all dependencies is a time-consuming and error-prone process. It lacks the managed features, scalability, and cost optimization benefits provided by Vertex AI Training for large-scale ML workloads.

Concept tested: Scalable and cost-effective ML training on Vertex AI

Source: https://cloud.google.com/vertex-ai/docs/training/create-custom-job

Topics

#ML Training#Vertex AI Custom Training#GPU Acceleration#Cloud Resource Optimization

Community Discussion

No community discussion yet for this question.

Sign up to join the discussion
Full PROFESSIONAL-MACHINE-LEARNING-ENGINEER Practice