A Specialist has a very large training dataset in S3 (millions of records). They want to avoid copying all data to a SageMaker notebook's 5 GB EBS volume but still train on the complete dataset. Which workflow allows using the full dataset for training with minimal local copying?

The correct workflow involves validating code with a small local subset and then leveraging SageMaker's Pipe input mode for full dataset training. SageMaker's Pipe mode streams data directly from S3 to the training instance, avoiding the need to download the entire dataset to the local EBS volume, which is limited to 5 GB for notebooks. Other options are less efficient or practical. Launching an EC2 instance to attach S3 is an unnecessary manual step when SageMaker handles S3 integration directly. Using AWS Glue for testing training is not its primary purpose; Glue is for ETL. Training on a local EC2 instance after validating in SageMaker also bypasses SageMaker's managed training capabilities.