In Amazon SageMaker, which mode allows for streaming training data directly from Amazon S3?

In Amazon SageMaker, the mode that allows for streaming training data directly from Amazon S3 is the Pipe mode. This mode is specifically designed to handle large datasets by streaming data from an S3 bucket directly into the training algorithm without the need to load all the data into local storage beforehand. It provides an efficient way to handle large-scale data, enabling real-time data processing as training occurs. This means that as the training process reads the data on-the-fly, it reduces latency and allows the algorithm to work with data that may not fit in memory.

In contrast, Batch mode is intended for scenarios where the training data is already available and can be fed to the algorithm in larger batches, whereas Real-time mode is typically associated with inference rather than training. File mode generally indicates that data is loaded in larger file chunks rather than streamed, making it less efficient for workflows requiring immediate data access during the training process.