Prepare and test your skills

Prepare and test your skills

Free sample question with answers and detailed explanation below. The Cloud Digital Leader practice bank has an additional 3,070 practice questions.Try 10 free realistic questions: instant score, no credit card, and no account required.
An enterprise is transitioning an open large language model (LLM) from an initial prototype to full production on Google Cloud. The engineering team has established the following technical and operational requirements for the deployment:
Which Google Cloud model serving approach should the enterprise choose?
Gemini Enterprise Agent Platform prebuilt container images are ready-to-use, Google-optimized container images configured for popular open-model serving frameworks—such as vLLM, Hex-LLM, SGLang, TGI, and TensorRT-LLM—that allow teams to self-deploy models without building or maintaining their own container images.
vLLM, TGI, and TensorRT-LLM out of the box.This option would be optimal if the enterprise wanted to self-deploy an open model on dedicated hardware for predictable, high-volume traffic, but wanted ease of use with an optimized serving framework like vLLM without having to build, customize, or maintain its own container images.
When self-deploying models on Google Cloud, teams choose between prebuilt containers and custom containers based on the trade-off between ease of use and maximum flexibility. Prebuilt containers are preferred unless custom dependencies or specialized pre/post-processing steps make a custom container necessary.
One-click deployment in Model Garden is a streamlined self-deployment feature that allows users to quickly provision and serve open models from Model Garden onto dedicated Google Cloud infrastructure using default, pre-configured serving settings.
This option would be appropriate if the engineering team needed a dedicated endpoint in their own Google Cloud project for security or hardware control, but wanted the fastest possible setup for a standard open model without building custom containers or writing custom pre-processing logic.
Model Garden acts as the central catalog for discovering and deploying open models in Google Cloud. While one-click deployment is ideal for getting a dedicated endpoint running quickly, workloads requiring deep container customization and custom serving logic must step up to a custom vLLM container deployment.
Model as a Service (MaaS) is a fully managed, serverless offering in Gemini Enterprise Agent Platform that exposes open large language models—such as Llama, DeepSeek, Mistral, and Qwen—through ready-to-use managed APIs without requiring users to provision or manage underlying infrastructure.
This approach would be the right choice if the organization wanted to minimize operational overhead, had unpredictable or bursty traffic patterns, and needed to rapidly prototype an application using a standard, non-customized foundation model.
Organizations frequently start with MaaS during the initial exploration and prototyping phases to validate business value quickly, and later transition to self-deployed models when traffic becomes high-volume and predictable or when custom fine-tuned weights are required.
Self-deploying a model with a custom vLLM container is an advanced deployment approach in Gemini Enterprise Agent Platform that allows organizations to build, configure, and manage their own serving container images while deploying open models on dedicated Google Cloud infrastructure.
While managed options like Model as a Service (MaaS) eliminate infrastructure management and prebuilt container images simplify standard deployments, neither allows modifying the underlying container dependencies or injecting unsupported pre-processing logic. Combining self-deployment with a custom vLLM container is the only option that satisfies the need for dedicated hardware selection, cost optimization at high volume, custom model weights, and custom container logic.
Answer them yourself and get your score the moment you finish. No account needed.
Full access includes
Keep the momentum going with these hand-picked practice scenarios
Last updated