Kvíz spojený s průvodcem · Střední úroveň

Základní ML pro kvíz o nasazení Apple na zařízení

Převeďte a integrujte model s Core ML při zohlednění typů balíčků, cílů nasazení, předběžného zpracování, výběru hardwaru a testování zařízení.

Související vodicí cestyCore ML On Device Deployment
Otázka 1 z 8

What role does Core ML play in an Apple app?

Náhled každé otázky8 otázky

Odpovědi a vysvětlení se zobrazují, jak budete v kvízu.

  1. What role does Core ML play in an Apple app?
    • Creating model architectures and optimizing their training loss inside an app
    • Converting supported source models for use by a Core ML app
    • Running trained-model inference through supported Apple software and hardware
    • Transforming source checkpoints into Core ML artifacts without separate conversion tools
  2. What does Core ML Tools commonly do in a conversion workflow?
    • Adjusts a source model's class labels to match the app
    • Converts supported source models into Core ML representations
    • Selects runtime compute units for each prediction
    • Executes the compiled model inside the app
  3. Why set and check a minimum deployment target?
    • It selects a training optimizer for the model
    • It controls which cloud region stores the model
    • It chooses an accelerator at runtime
    • Model formats and APIs have minimum OS availability requirements
  4. What can cause Core ML predictions to differ from the original model?
    • Different preprocessing, such as color order or normalization
    • A renamed model resource in the app bundle
    • A changed compute-unit preference
    • A different source-code build number
  5. Does a compute-unit setting guarantee each operation runs on the Neural Engine?
    • Yes, a compute-unit preference guarantees each operation uses the Neural Engine
    • Yes, the setting chooses the fastest accelerator for every device
    • No; compute-unit settings constrain allowed resources, but placement depends on device and operation support
    • No; Core ML executes only on CPU regardless of configuration
  6. Why benchmark on physical target devices?
    • Simulator timing captures thermal throttling and memory pressure on every target device
    • Latency, memory, and energy depend on actual hardware and OS
    • All Apple device families use an identical execution plan
    • Every physical device uses the same supported model formats regardless of OS version
  7. What should be checked when converting to an ML Program model?
    • That it always uses the legacy single-file model format
    • That conversion changes it into a TensorFlow checkpoint
    • That a converted model requires a network service
    • Its package format and minimum OS availability
  8. Which practice supports a reliable on-device prediction path?
    • Compare app outputs after integrating preprocessing and model loading
    • Assume conversion success proves equivalent predictions
    • Test the artifact only in a simulator
    • Skip input validation because the model declares its shape