Skip to main content

ML models tab

Open Assessments > Drift & Bias > ML models to see every classic model on one list.

Figure 1. The ML models tab: actions, headline counts and one row per model.

Actions​

ButtonWhat it does
ExportOpens a new ML model monitoring export in Export Center. See Export.
Add modelShows both ways to connect a model: from your pipeline or from MLflow.
Check allRuns a check on every model now. See How checks run.

Headline counts​

CardCounts
ML modelsClassic models known from a pipeline push or MLflow. LLMs are tested by runs instead.
DriftingModels with data or prediction drift in the latest check.
DegradedModels whose primary metric worsened past its tolerance.
Fairness flaggedModels where a group fell outside the fairness limits.
Open findingsOpen findings across every model.

The model table​

ColumnShows
ModelName and model key. Select a model to open its model page.
SourcePipeline, MLflow or Pipeline + MLflow.
TaskFor example Binary classification or Regression.
Latest versionThe newest version, and how many versions are tracked.
Last checkedWhen the latest check ran, or Not yet.
DriftTwo badges: Data and Prediction, each Stable, Moderate, Severe or Not checked.
PerformanceHolding, Degraded or No labels.
FairnessFlagged, or Not checked.
Open findingsThe model's open findings.

The drift levels come from your PSI limits; Degraded comes from the performance tolerance; Flagged comes from the fairness limits. All three are set per model in Monitoring settings.