Seems like I didn't test the option to save results, late interaction models and cross encoders for reranking.