引導式測驗 · 中等 等級

GitHub ML 管道測驗的操作

透過操作作業、工件可追溯性、受保護的發布以及對運行程序和憑證的仔細處理,建立分階段的 ML 自動化。

相關引導路徑ML 管道的 Github 操作
問題 1 的 8

GitHub 操作工作流程定義什麼?

預覽每個問題8 問題

當您參加測驗時,就會出現答案和解釋。

  1. GitHub 操作工作流程定義什麼?
    • Triggers, jobs, runners and steps
    • A container image manifest and layer list
    • Branch protection rules for a repository
    • The model dependency lockfile alone
  2. 為什麼將快速拉取請求檢查與完整的 GPU 訓練分開?
    • It guarantees model quality without evaluation
    • It keeps feedback quick while reserving specialized compute for expensive work
    • It removes all runner maintenance
    • It makes GPU hardware available on every runner
  3. 評估報告與之後推廣的工件應該如何連結?
    • The author’s username alone
    • The workflow display name only
    • The artifact digest or another immutable identity
    • A cache key that changes each run
  4. 為什麼要避免將儲存庫機密暴露給不受信任的拉取請求程式碼?
    • Actions cannot run without write tokens
    • Secrets are required to run Python imports
    • It makes all validation nondeterministic
    • Workflow code could use them to access external systems
  5. 與雲端提供者配置後,OIDC 可以提供什麼?
    • A permanent model accuracy certificate
    • Short-lived credentials tied to workflow identity
    • An artifact cache without access control
    • A Python virtual environment
  6. How should a dependency cache be used in an Actions workflow?
    • Keep a deployable model file between workflow runs as the official release record
    • Reuse dependencies with a key that ignores dependency input changes
    • Speed up repeated dependency setup with keys based on dependency inputs
    • Use a mutable image tag as the only record of workflow output
  7. What does an environment protection rule add to a model release job?
    • A lockfile for training data
    • An automatic guarantee that metrics are unbiased
    • A new GPU device
    • A controlled approval or deployment condition
  8. What does GitHub Actions alone not guarantee?
    • That training data, evaluation or release decisions are sound
    • That defined workflow steps execute on a runner
    • That jobs can depend on prior jobs
    • That artifacts can be uploaded