Kvíz spojený s průvodcem · Střední úroveň

Kvíz Amazon SageMaker pro pracovní postupy ML

Propojte úlohy školení SageMaker, koncové body, kanály a správu modelů při zohlednění oprávnění, hodnocení, kvót a aktuálních nákladů.

Související vodicí cestyPracovní postupy Amazon Sagemaker ML
Otázka 1 z 8

Which SageMaker workload executes configured model training on managed compute?

Náhled každé otázky8 otázky

Odpovědi a vysvětlení se zobrazují, jak budete v kvízu.

  1. Which SageMaker workload executes configured model training on managed compute?
    • Hosting online predictions from an already trained artifact
    • Transforming a static dataset with a deployed model
    • Running configured model training on managed compute
    • Recording metadata for a model version without running training
  2. Which SageMaker resource is intended for real-time model requests?
    • A batch transform job
    • A hosted endpoint
    • A pipeline definition
    • A training checkpoint
  3. Why may a hosted endpoint cost money when request volume is low?
    • An endpoint automatically trains the model between requests
    • Every endpoint stores an entire copy of its training dataset
    • A prediction triggers a separate training job
    • Provisioned serving capacity may remain active while waiting for requests
  4. What can a pipeline automate in an ML workflow?
    • Connecting repeatable steps such as preprocessing, training, and evaluation
    • Choosing data labels without a specified annotation policy
    • Evaluating model quality without defining any metrics
    • Removing the need to configure permissions for jobs
  5. What should be reviewed before estimating SageMaker cost?
    • Only model parameter count
    • Only training accuracy
    • Region, instance type, duration, storage, and serving configuration
    • Only whether the model uses a built-in algorithm
  6. How should a model package move toward deployment?
    • Assume registration itself is production approval
    • Evaluate and review against explicit criteria before release
    • Deploy every training output automatically
    • Use test data repeatedly for tuning after evaluation
  7. What does least-privilege IAM accomplish in a training job?
    • It grants every training container administrator access
    • It changes the model's inference behavior
    • It bypasses object-storage permissions
    • It restricts a job to the data and artifacts it needs
  8. What should happen after a temporary experiment ends?
    • Verify temporary endpoints, compute, and storage are cleaned up
    • Keep idle endpoints active without checking their configuration
    • Delete model lineage while retaining all temporary compute
    • Stop the job but leave temporary volumes and artifacts unreviewed