2.4 Leveraging Gemini Enterprise Agent Platform for End-To-End ML Workflows PROFESSIONAL-CLOUD-ARCHITECT Practice Questions (Page 1)

Questions 1–5

1. 1foundation · easy

   Which consumption model is best suited for a predictable, long-running training job that must not be interrupted?

   • AOn-demand VMs
   • BSpot VMs
   • CCommitted use discounts
   • DPreemptible VMs
   Reveal Solution Discussion Select an answer first
2. 2application · medium

   A financial services firm needs to automate its model retraining workflow. The workflow involves data validation, feature engineering, model training, and evaluation. They want to ensure that each step is reproducible and that the entire process can be triggered automatically when new data arrives. What should they use?

   • AAgent Platform Pipelines with a Cloud Storage trigger that starts the pipeline on new data upload.
   • BA series of Cloud Functions that call each other sequentially, triggered by a Cloud Scheduler job.
   • CA single Cloud Run service that performs all steps (validation, feature engineering, training, evaluation) in one container.
   • DA Vertex AI TensorBoard instance to monitor the training and trigger the next step.
   Reveal Solution Discussion Select an answer first
4. 3application · medium

   A team is building a pipeline that includes a step to run a custom container on Cloud Run. They want to pass the output of a previous pipeline step (a file path) to this Cloud Run step. How should they pass this data?

   • AUse a pipeline parameter to pass the file path as a string to the Cloud Run step.
   • BHardcode the file path in the Cloud Run container's environment variables.
   • CHave the Cloud Run step read the file path from a fixed location in Cloud Storage.
   • DUse a Vertex AI ML Metadata store to pass the file path.
   Reveal Solution Discussion Select an answer first
5. 4expert · hard

   A team is running a training job on AI Hypercomputer and is experiencing performance issues. They suspect that the network bandwidth between the nodes is the bottleneck. What is the most effective way to diagnose and address this?

   • AIncrease the number of nodes in the cluster to distribute the load.
   • BUse a higher-bandwidth interconnect, such as a larger TPU slice or a GPU cluster with a faster network, and profile the job to confirm the bottleneck.
   • CSwitch to a different model architecture that requires less communication.
   • DReduce the batch size to decrease the amount of data transferred.
   Reveal Solution Discussion Select an answer first
6. 5application · medium

   A retail company wants to build a pipeline that trains a recommendation model. Their data is spread across BigQuery, Cloud Storage (CSV files), and an on-premises Oracle database. They plan to use Agent Platform Pipelines to orchestrate the workflow. What is the first step they should take to prepare their data for integration?

   • AWrite a custom pipeline component that connects to each source (BigQuery, GCS, Oracle) and joins the data in memory.
   • BConvert all data to CSV files and store them in a single Cloud Storage bucket before starting the pipeline.
   • CCreate a unified dataset in BigQuery by loading the CSV files and replicating the Oracle data using a batch transfer job.
   • DUse a Dataflow job within the pipeline to read from all three sources and write the combined result to a temporary location.
   Reveal Solution Discussion Select an answer first