2.4 Leveraging Gemini Enterprise Agent Platform for End-To-End ML Workflows PROFESSIONAL-CLOUD-ARCHITECT Practice Questions (Page 5)

Questions 21–25

1. 21foundation · easy

   What is a key consideration when using spot VMs for ML workloads on Agent Platform?

   • ASpot VMs are guaranteed to run for at least 24 hours.
   • BSpot VMs can be terminated at any time, so workloads must be fault-tolerant and checkpointed.
   • CSpot VMs provide the same performance as regular VMs but at a higher cost.
   • DSpot VMs are only available for serving workloads, not training.
   Reveal Solution Discussion Select an answer first
2. 22application · medium

   A media company runs a nightly pipeline that ingests video metadata from Cloud Storage, transforms it using a custom Python function, and then triggers a Vertex AI training job. The team wants to add a step that calls an external sentiment-analysis API hosted on Cloud Run, but only if the transformation step produces more than 100 rows. How should the pipeline be configured to meet this requirement?

   • AModify the transformation function to call the Cloud Run API directly when the row count exceeds 100, and keep the pipeline as a single step.
   • BAdd the Cloud Run call as a separate pipeline step and use a conditional (if/else) gate on the output row count from the transformation step.
   • CCreate a second pipeline that runs after the first completes, and have it check the output row count before calling the Cloud Run service.
   • DUse a Cloud Scheduler job that runs every hour, checks the latest output, and invokes the Cloud Run service if the row count is above 100.
   Reveal Solution Discussion Select an answer first
4. 23application · medium

   A team has a pipeline that trains a model. After training, they want to run a lightweight evaluation step that is implemented as a Cloud Run function. The evaluation step needs to access the trained model artifact. How should they pass the model artifact to the Cloud Run function?

   • AHave the Cloud Run function download the model from a public URL.
   • BPass the Cloud Storage URI of the model artifact as a pipeline parameter to the Cloud Run step.
   • CEmbed the model artifact in the Cloud Run function's container image.
   • DUse a Vertex AI Model Registry resource to pass the model.
   Reveal Solution Discussion Select an answer first
5. 24expert · hard

   A company is training a very large model that requires more memory than a single TPU or GPU device can provide. They are considering using model parallelism across multiple devices. What is the key consideration when using model parallelism on AI Hypercomputer?

   • AModel parallelism is automatically handled by the framework and requires no manual configuration.
   • BModel parallelism is only supported on TPUs, not GPUs.
   • CModel parallelism eliminates the need for data parallelism.
   • DThe communication overhead between devices can become a bottleneck, so the model must be partitioned to minimize cross-device communication.
   Reveal Solution Discussion Select an answer first
6. 25foundation · easy

   What is a key consideration when preparing data integration with Agent Platform?

   • AEnsuring data is stored in a location accessible to the pipeline, such as Cloud Storage or BigQuery.
   • BEnsuring data is encrypted only during the training step.
   • CConverting all data to a single proprietary format before integration.
   • DManually copying data to every VM that will run a pipeline component.
   Reveal Solution Discussion Select an answer first