Hello! Did you try specify a column name as manual split column?
Then you can fully control which rows are in train / validate / test splits: https://docs.databricks.com/en/machine-learning/automl/automl-data-preparation.html#split-data-for-regressi...
