capabilities.work

Serving Route Inference by Version Label

A 3-step recipe with TensorFlow. We can help you get access.

Recipe · 3 steps

Serving Route Inference by Version Label

with TensorFlow · Artificial Intelligence Management

Resolve a version label to a version number and predict against that pinned version.

  1. Resolve the label to the model version currently behind it and confirm that version is AVAILABLE. The resolved version number is captured as an output so the rest of the flow is pinned and the run stays reproducible even if the label is mov
  2. Read the signature metadata of the version the label resolved to, so the caller knows exactly which model shape it is about to send inputs to.
  3. Submit the prediction pinned to the version number resolved from the label, rather than to the label itself, so every instance in this request is scored by the same model.
What you'll need to have ready
  • modelName: The name the model was registered under in the ModelServer.
  • label: The version label to route by, as configured on the ModelServer (commonly "stable" or "canary").
  • signatureName: Optional signature to target. When omitted the model's default serving signature is used.
  • instances: Row format prediction inputs, one prediction returned per instance.
For your developer

Arazzo workflow tensorflow:label-routed-inference, 3 operations: getModelLabelStatus → getModelVersionMetadata → predictModelVersion. Specs on APIs.io.

More recipes with TensorFlow

Serving Preflight and ClassifyServing Pinned Reproducible Example ScoringServing Preflight and PredictServing Preflight and RegressServing Gate a Rollout on Version ReadinessServing Compare a Candidate Version Against the Default