Kvíz spojený s průvodcem · Střední úroveň

Kvíz Google Vertex AI Platform

Sledujte data Vertex AI, školení, kanály, registr a pracovní toky koncových bodů a zároveň kontrolujte hodnocení, IAM, monitorování, kvóty a náklady.

Související vodicí cestyPlatforma Google Vertex AI
Otázka 1 z 8

Which workflow component, called Vertex AI Pipelines in established docs, can orchestrate repeatable preprocessing and training steps?

Náhled každé otázky8 otázky

Odpovědi a vysvětlení se zobrazují, jak budete v kvízu.

  1. Which workflow component, called Vertex AI Pipelines in established docs, can orchestrate repeatable preprocessing and training steps?
    • An endpoint that serves online predictions
    • Cloud Logging by itself
    • Vertex AI Pipelines
    • A service-account role
  2. What does deploying a model to an endpoint enable?
    • Automatic retraining from each user request
    • Online inference requests
    • Version control of training source code only
    • Granting the endpoint access to all project storage
  3. What does a model registry entry establish?
    • A guarantee of production safety
    • A replacement for evaluation data
    • An endpoint already serving traffic
    • A managed model-version record, not automatic deployment approval
  4. Why review pipeline caching and artifact identity?
    • Incorrect cache reuse can pass stale outputs into a later run
    • A cache always invalidates when upstream data changes
    • Pipeline caching guarantees every component is deterministic
    • Caching removes the need to identify code and input versions
  5. Which cloud permissions and settings affect Vertex AI execution?
    • The model's objective function alone
    • The selected machine accelerator alone
    • Service-account roles, network configuration, and artifact access
    • The model's prediction threshold
  6. What should be checked before estimating Vertex AI costs?
    • Only the count of pipeline components
    • Current region, machine type, endpoint duration, and managed services
    • Only batch size and model accuracy
    • Only which SDK created the model
  7. How should endpoint monitoring be designed?
    • Log all raw sensitive user inputs by default
    • Use training accuracy as the only production signal
    • Assume deployment itself validates the model
    • Track latency and errors plus appropriate quality or input signals
  8. When does Vertex AI provide a practical platform fit for an organization?
    • The workflow fits managed Google Cloud integration and the organization's constraints
    • The platform guarantees a model generalizes to new data
    • The service removes the need for security review
    • Every workload costs less when managed