QuestionQ15

VerifiedDeployment and Orchestration of ML Workflows

A company has an existing Amazon SageMaker model (v1) on a production endpoint. The company develops a new model version (v2) and must test v2 in production before replacing v1 with v2.

The company must implement a solution that minimizes the risk of v2 producing incorrect output in production. The solution must prevent any disruption to production traffic while changing to v2.

Which solution meets these requirements?

Explanation

An Amazon SageMaker shadow variant receives replicated live inference requests, but its responses are not returned to callers; v1 continues to receive and respond to production traffic. Sampling 100% of requests lets v2 be assessed against the full production workload, and the logged outputs can be evaluated before promoting v2 to the production variant.

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!