引導式測驗 · 中等 等級
ML 服務測驗的 SLO 和警報
定義 ML 服務指標和目標,選擇與使用者影響相符的警報,並將營運 SLO 與模型品質目標區分開來。
問題 1 的 8
What does an SLO specify?
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。
- What does an SLO specify?
- A target for a measured service indicator over a defined period
- A model's training objective
- A list of every possible alert
- A guarantee that all requests are identical
- Why might a latency SLO use a percentile instead of the mean?
- It removes all variation in request time
- It captures slow-tail requests that an average can hide
- It measures model accuracy
- It guarantees zero timeouts
- Which indicator may suit a batch scoring service?
- Training accuracy only
- Per-request p99 only
- Completion by deadline and feature-data freshness
- 模型參數數量
- What can an error budget represent?
- A guarantee that no incident will occur
- The number of incorrect model labels in training
- The cloud provider's monthly bill
- Allowed unreliability under a defined SLO
- Why separate service availability from model quality monitoring?
- Both metrics always move together
- A service can respond successfully while producing poor predictions
- Quality can be measured only with latency
- Availability is a model parameter
- Which condition is most suitable for an urgent page?
- A minor dashboard formatting change
- A small one-time shift in unlabeled feature distribution
- Sustained errors or capacity exhaustion affecting users
- A model's parameter count increased
- Why define eligible requests in an availability SLI?
- It changes the model's predictions
- It makes every failed request successful
- It removes the need for monitoring windows
- The denominator must specify which events count toward success
- What makes an alert useful beyond a metric dashboard?
- A threshold, response owner and actionable runbook
- A chart with many colors
- An unbounded list of every feature
- A model checkpoint filename