An ML engineer has deployed an Amazon SageMaker model to a serverless endpoint in production. The model is invoked by the InvokeEndpoint API operation. The model's latency in production is higher than

Question

Options

• ASchedule a SageMaker Model Monitor job. Observe metrics about model quality.
• BSchedule a SageMaker Model Monitor job with Amazon CloudWatch metrics enabled.
• CEnable Amazon CloudWatch metrics. Observe the ModelSetupTime metric in the SageMaker
• DEnable Amazon CloudWatch metrics. Observe the ModelLoadingWaitTime metric in the

• A
  6% (2)
• B
  3% (1)
• C
  15% (5)
• D
  76% (26)

Explanation

Option D is correct because ModelLoadingWaitTime is the specific Amazon CloudWatch metric that measures how long a SageMaker serverless endpoint spends loading the model container before it can serve a request - this is precisely the "cold start" latency the engineer suspects, and enabling CloudWatch metrics exposes this value directly.

Option C is wrong because ModelSetupTime is not a valid SageMaker CloudWatch metric; the actual metric for model loading duration on serverless endpoints is ModelLoadingWaitTime, so C uses a fabricated or incorrect metric name.

Options A and B are wrong because SageMaker Model Monitor is purpose-built to track data quality, model quality, bias drift, and feature attribution drift - not endpoint performance or startup latency. Adding CloudWatch metrics to a Model Monitor job (B) still does not expose model loading time.

Memory tip: Think of it as "waiting for the model to load = ModelLoadingWaitTime." Serverless endpoints have cold starts; this metric is your cold-start clock. Model Monitor = data/quality drift; CloudWatch metrics on the endpoint = performance diagnostics.

No community discussion yet for this question.