引導式測驗 · 中等 等級

使用 FastAPI 測驗服務模型

設計經過驗證的 FastAPI 預測端點,並將 API 功能與模型載入、並發和生產控制區分開來。

相關引導路徑使用 Fastapi 服務模型
問題 1 的 8

What can typed request models provide in FastAPI?

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. What can typed request models provide in FastAPI?
    • Input validation and a documented API schema
    • Guaranteed correct model predictions
    • GPU driver installation
    • Automatic model retraining
  2. When should a large model artifact usually be loaded?
    • Inside every request handler call
    • During application startup and reused across requests
    • After sending the HTTP response
    • Only when the process shuts down
  3. Why can synchronous CPU inference inside an async endpoint be problematic?
    • It disables input validation
    • It converts JSON into GPU memory automatically
    • It can block the event loop and delay other requests
    • It makes predictions deterministic
  4. What risk can multiple worker processes create for a GPU model?
    • They disable model versioning
    • They combine all copies into one automatically
    • They remove GPU memory limits
    • Each process may load another copy of model weights into memory
  5. What should readiness indicate for a model API?
    • Whether the training set is alphabetized
    • Whether the model and required dependencies are ready to serve
    • Whether every prediction is correct
    • Whether the server source code has comments
  6. Why add timeouts and concurrency limits?
    • To make all requests asynchronous
    • To increase the model's accuracy
    • To bound resource use and prevent overload from unbounded requests
    • To remove authentication requirements
  7. What does OpenAPI documentation not provide by itself?
    • A list of API operations
    • A description of request fields
    • A schema for responses
    • Authentication or authorization enforcement
  8. When might a dedicated model server be preferable?
    • Measured throughput, batching or model-management needs exceed a simple API process
    • Whenever the model has one input
    • Before measuring request load
    • When input validation is required