Skip to content

6.9.2.8 ModelVersion

6.9.2.8.1 Purpose

One trained version of a model, produced by a training run and kept as an artifact in the model registry. Deploying a version makes it the one that answers predictions and retires the previous one; a retired version refuses calls.

6.9.2.8.2 Questions it answers

  • Which training run produced the version that is deployed?
  • When was a version deployed, and when retired?
  • Which artifact and endpoint serve it?

6.9.2.8.3 Fields

Field Type Cardinality Description
id string (uuid) 1..1 Key.
model Model 1..1
version string 1..1 Semver label. (model, version) is unique.
training_run ModelRun 1..1
artifact_path string 1..1 The artifact in the model registry.
endpoint string 0..1 The Vertex AI endpoint or the batch pipeline that serves it.
status enum 1..1 Values: registered, deployed, retired. Default: registered.
Deploying a version retires the one deployed before it; a retired version refuses calls.
deployed_at datetime 0..1
retired_at datetime 0..1
created_by MoonitoringUser 1..1
created_at datetime 1..1
updated_at datetime 1..1

6.9.2.8.4 Relationships

References

Referenced by

6.9.2.8.5 Rules & constraints

  • One trained version of a model, from a training run. Its content never changes; only its status does.