Skip to main content

Export ML model checks

The ML model monitoring source in Export Center lists the checks of your classic ML models started in the selected period, one row per check, with its drift, performance and fairness outcome.

note

Exports hold aggregates only. Pushed rows, predictions and labels are not included.

Create the export​

Figure 1. Creating an ML model monitoring export.

  1. Select Export on the ML models or Baselines tab. Export Center opens Create ML model monitoring export with ML model monitoring already chosen as the Source. You can also start from Settings > Data management > Export Center > Create export.
  2. Work through the steps. The Draft manifest on the right summarizes each one:
    • 01 Source: Export name and Source.
    • 02 Query: the Period (for example Last 30 days), Search and Conditions. Records must match every condition; OR, NOT and nested groups are not supported.
    • 03 Fields: choose the fields and arrange their Output order.
    • 04 Delivery: format and schedule.
    • 05 Review: preview the draft.
  3. Select Create export. It is enabled once you preview the current draft.

See Export Center for delivery formats, schedules and limits.

Export LLM runs​

The Drift & Bias runs source exports one row per LLM run: its model, question set, status, bias-free score and interval, baseline comparison, drift result and each group's results. Filter by status, question set, model, baseline or drift. Replies to the test questions are never exported.