Optionaltop_column_scores?: Record<string, number>[] | nullColumn scores per query row extracted from the model (higher means more weight was put on this column).
Optionaltop_relevant_context_rows?: number[][] | null2D array where each subarray contains indices of most relevant context rows for that query row. The first dimension indexes query rows, the second dimension indexes all rows as a sequential integer index.
Explanation data for predictions.