Which Amazon SageMaker inference option is appropriate for performing inference on large archived datasets of multiple GBs in size when real-time access to predictions is not required?