Skip to main content

Overview tab

The Overview tab keeps LLMs and classic ML models apart, in two numbered sections.

Figure 1. The Overview tab: 01 Text models (LLMs) on top, 02 Classic models (ML models) below.

01 Text models (LLMs)​

How your text models treat gender, race and ethnicity, and age, one question set at a time.

  • Question set: choose which question set to show, then switch between Bias-free score, Accuracy and Refusal rate.
  • Headline counts: Series (model, question set, judge and settings that are compared), Drifted (series whose latest run differs from its baseline), Lowest group and Widest gap (between two groups of one kind).
  • Bias-free score by model and group: one row per series, showing its latest finished run, worst group first. Groups are arranged under Gender, Race & ethnicity and Age, with Trend and Latest. The stronger the colour, the worse the score. A cell with fewer than 30 usable replies is marked as such. Select a row or a cell to open its run. Switch between Chart and Table.
  • All LLM runs opens the LLM runs tab.
  • Live bias, 7 days: when Live bias check is on for an app in the LLM Gateway, the tab shows the live rate across your apps and, under each model, its live rate next to its test-run score.

02 Classic models (ML models)​

Your classic ML models, checked on their own scored rows for drift, performance and fairness. The headline counts are the same as on the ML models tab: ML models, Drifting, Degraded, Fairness flagged and Open findings.

Select All ML models to open the ML models tab.