sample data and prerun model weight