26.07.02 Release notes

This release adds font color controls to the rich text editor, enables built-in DeepEval scorers to use a configured judge model, and fixes MinimumF1Score for multiclass data with missing true labels.

Enhancements

Frontend

Set text and background colors in the rich text editor

You can now set the font color and background color of text in rich text content blocks.

ValidMind Library v2.13.6

Configure the judge model used by built-in DeepEval scorers

Built-in DeepEval scorers such as AnswerRelevancy and Hallucination now use the judge model set with set_judge_config(). This lets you use supported providers, including Vertex AI, without configuring a separate model for each scorer.

Bug fixes

ValidMind Library v2.13.6

Fixed error in MinimumF1Score for multiclass detection with missing true labels

Fixed an issue where MinimumF1Score raised an error stating "Target is multiclass but average='binary'" when a model predicted a class not present in a dataset split's true labels. Now, the metric handles these situations without errors.

Fixed failure in IQROutliersBarPlot with boolean feature columns

Fixed a failure in IQROutliersBarPlot when datasets included boolean feature columns. Now, boolean and binary features are skipped, preventing the numpy boolean subtract error during IQR calculations.