QuestionQ88

VerifiedServing and scaling models

You developed an ML model that detects sentiment in users’ posts on your company’s social-media page to identify outages or bugs. You use Dataflow to deliver real-time predictions for data ingested from Pub/Sub. You plan to train the model through multiple iterations and retain the newest two versions live after each run. You want traffic divided 80:20 between the versions, with the newest model receiving most traffic. You also want the simplest pipeline possible with minimal management. What should you do?

Explanation

Vertex AI endpoints can serve multiple deployed models and route prediction traffic by percentage. In a deployment traffic split, 0 represents the newly deployed model, so 0=80, PREVIOUS_MODEL_ID=20 directs 80% of requests to the newest model while retaining the prior model for the remaining 20%.

Learn more

Community Discussion

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