Kuiz berkaitan panduan · Sederhana Tahap

Pengawas dan Dagster untuk Kuiz Orkestrasi ML

Bandingkan aliran Pengawas dan aset Dagster untuk kerja ML, termasuk percubaan semula, keturunan, penjadualan, penempatan dan kesesuaian operasi.

Laluan panduan yang berkaitanPengawas Dan Dagster Untuk ML
soalan 1 daripada 8

How does Prefect commonly define a flow?

Pratonton setiap soalan8 soalan

Jawapan dan penjelasan muncul semasa anda mengambil kuiz.

  1. How does Prefect commonly define a flow?
    • A Python function that cannot call another flow
    • A model evaluation metric that schedules jobs
    • A decorated Python function that composes workflow logic
    • A graph of named data assets and their dependencies
  2. How does Dagster represent an asset-oriented workflow?
    • An invocation record for a scheduled workflow
    • Named data outputs and their production dependencies
    • A task execution state with no output lineage
    • A set of training hyperparameters
  3. Why can an automatic retry be unsafe for some workflow steps?
    • Prefect automatically detects non-idempotent operations and suppresses duplicate writes
    • A task retry never calls its Python function more than once
    • Setting a retry count of one prevents the first attempt from running
    • A retry can repeat an external side effect when the operation is not idempotent
  4. What should be recorded to support ML artifact lineage?
    • Dataset identity, code revision, configuration, and artifact version
    • Only the schedule and run name
    • Only the UI layout used to launch jobs
    • Hanya markah ketepatan akhir
  5. What should teams compare in a representative orchestrator trial?
    • Only whether each workflow step is decorated
    • Only the preferred dashboard color scheme
    • Debugging, deployment, lineage, backfills, and operating requirements
    • Only validation accuracy for the trained model
  6. When might Dagster's asset model be useful?
    • When a task has no outputs to track
    • When teams need to track produced data and model outputs with dependencies
    • Only when serving models on phones
    • When the workflow should have no state
  7. What should happen to a training job that needs specialized compute?
    • The orchestrator should run all GPU training inside its own process
    • The training job should be hidden inside a schedule definition
    • The model data should be stored as a secret
    • Submit the job to suitable compute and track its outputs
  8. Which statement about Prefect and Dagster is accurate?
    • Choose based on workflow abstractions and team operating needs
    • One always runs faster for every ML workflow
    • Either tool guarantees model validity
    • Both remove the need to track data versions