2.4 Leveraging Gemini Enterprise Agent Platform for End-To-End ML Workflows PROFESSIONAL-CLOUD-ARCHITECT Practice Questions (Page 3)

Questions 11–15

1. 11foundation · easy

   Why is checkpointing important for large-scale AI model training?

   • AIt automatically improves model accuracy.
   • BIt eliminates the need for distributed training.
   • CIt allows the training job to resume from a saved state if it fails or is interrupted.
   • DIt reduces the amount of training data needed.
   Reveal Solution Discussion Select an answer first
2. 12expert · hard

   A company is planning a large-scale training job that will run for several weeks. They want to minimize cost but also want to ensure the job makes steady progress. They are considering using spot instances with checkpointing. What is the most important trade-off to manage?

   • AThe number of spot instances versus the number of on-demand instances.
   • BThe frequency of checkpointing versus the cost of saving checkpoints and the time lost on preemption.
   • CThe size of the model versus the size of the dataset.
   • DThe choice between using GPUs and TPUs.
   Reveal Solution Discussion Select an answer first
4. 13foundation · easy

   In Agent Platform Pipelines, what is a 'component'?

   • AA dataset that is used as input to the pipeline.
   • BA container image that stores the final trained model.
   • CA self-contained step in the pipeline that performs a specific task, such as data preprocessing or model training.
   • DA virtual machine that runs the entire pipeline.
   Reveal Solution Discussion Select an answer first
5. 14application · medium

   A healthcare company needs to train a model on sensitive patient data. They have strict data residency requirements and must ensure the data does not leave a specific region. They plan to use Agent Platform Pipelines. What is the most important consideration for data integration?

   • AUse a Private Google Access connection for all data transfers.
   • BEnsure all data is stored and processed in a Google Cloud region that meets the residency requirement.
   • CSet up a VPC Service Controls perimeter to restrict data access.
   • DUse customer-managed encryption keys (CMEK) for all data at rest.
   Reveal Solution Discussion Select an answer first
6. 15application · medium

   A data science team is training a large language model and needs to reduce the training time. They are currently using a single GPU node. They want to leverage the full capabilities of AI Hypercomputer. What is the most effective way to scale up their training?

   • AIncrease the memory of the single GPU node to allow for a larger batch size.
   • BSwitch to a single TPU v4 device.
   • CUse a single GPU with a higher clock speed.
   • DMove to a multi-host TPU v5e slice configuration and use a data-parallel training strategy.
   Reveal Solution Discussion Select an answer first