Kvíz spojený s průvodcem · Střední úroveň

Kvíz o podávání TensorFlow

Rozumějte signaturám SavedModel, zásadám verzí a dávkování v TensorFlow Serving a rozlišujte stav běhu od kvality modelu.

Související vodicí cestyPodání Tensorflow
Otázka 1 z 8

What does a SavedModel signature describe for a serving client?

Náhled každé otázky8 otázky

Odpovědi a vysvětlení se zobrazují, jak budete v kvízu.

  1. What does a SavedModel signature describe for a serving client?
    • Named input and output tensor contracts
    • The host's network firewall
    • The model's training objective only
    • The deployment's traffic percentage
  2. Why use versioned model directories?
    • To store every inference request
    • To make model versions available under serving configuration
    • To replace evaluation gates
    • To automatically compare fairness
  3. What does a version policy not establish?
    • How versions are selected by configuration
    • Which model versions may be served
    • That a candidate passed quality validation
    • That the server can load a model
  4. How can request batching affect serving performance?
    • It eliminates concurrency
    • Latency always decreases with batch size
    • Batching changes the SavedModel signature
    • Throughput can improve while queuing latency and memory use increase
  5. What can cause a serving request to fail despite a loaded model?
    • The model has a version directory
    • Input tensor names, dtypes or shapes do not match the signature
    • A REST endpoint exists
    • The client used TLS
  6. What does a model health endpoint generally indicate?
    • That labels have matured
    • That every response is correct
    • Process or model availability, not predictive accuracy
    • That fairness is proven
  7. Why might model warmup be configured?
    • To alter tensor names
    • To increase the number of model versions
    • To update training labels
    • To reduce first-request latency after loading
  8. What should be tested when changing a SavedModel version?
    • Signature compatibility and predictions in the target serving environment
    • Only directory naming
    • Only the model's training loss
    • Whether the server runs on a developer laptop