Bài kiểm tra có liên kết hướng dẫn · Trung bình Cấp độ

Bài kiểm tra chọn kích thước mẫu phù hợp

Chọn các mô hình bằng cách sử dụng kiểm tra chất lượng, độ trễ, chi phí, độ an toàn và phiên bản dành riêng cho nhiệm vụ thay vì giả sử một kích thước mô hình phù hợp với mọi yêu cầu.

Câu hỏi 1 của 8

How should a team select a model tier for a production task?

Xem trước mọi câu hỏi8 câu hỏi

Câu trả lời và giải thích xuất hiện khi bạn làm bài kiểm tra.

  1. How should a team select a model tier for a production task?
    • Evaluate candidate models against the task’s quality, latency, cost, and safety requirements
    • Always choose the largest model
    • Always choose the cheapest model
    • Use model size as the only metric
  2. Why can a smaller model be a good choice for a simple task?
    • It is guaranteed to outperform larger models
    • It never makes errors
    • It may meet the needed quality with lower cost or latency
    • It removes the need for evaluation
  3. How can a team compare candidate models fairly?
    • Compare only marketing descriptions
    • Use representative held-out examples and measure quality, errors, latency, and cost
    • Run one prompt once
    • Compare parameter names only
  4. What risk applies to routing based on model confidence?
    • Confidence always equals correctness probability
    • Routing removes all errors
    • Confidence prevents model updates
    • Confidence may not be calibrated, so escalation thresholds need testing
  5. What information should a production team record about a chosen model?
    • Exact model identifier or version and the evaluation date
    • Only the display name
    • Only the API hostname
    • Nothing after deployment
  6. When might a request be escalated to a more capable model?
    • Whenever it has fewer than ten words
    • When the model is cheaper
    • Only after every request is already complete
    • When it meets tested criteria for complexity or uncertainty
  7. What should be monitored after changing model versions?
    • Only token count
    • Only the deployment timestamp
    • Task quality, error types, latency, cost, and failures
    • Only model name length
  8. Does a larger model automatically perform better for every request?
    • Yes, for every task
    • No; task fit and evaluation results determine whether added capability helps
    • Yes, if its context window is longer
    • No, because larger models cannot follow instructions