Boto3 create_algorithm does not support FastFile

[fcdl94]

Describe the issue

While in the documentation is written that TrainingChannels accepts "SupportedInputModes": ["File", "Pipe", "FastFile"] it throws a validation error if you try to use "FastFile"

For example try:

    result = client.create_algorithm(
        AlgorithmName=algorithm_name,
        AlgorithmDescription="Training_Algorithm",
        TrainingSpecification={
            "TrainingImage": train_image,
            "SupportedHyperParameters": hyperparameters,
            "SupportedTrainingInstanceTypes": supported_instances,
            "SupportsDistributedTraining": False,
            "MetricDefinitions": [],
            "TrainingChannels": [
                {
                    "Name": "dataset",
                    "Description": "full dataset channel",
                    "IsRequired": False,
                    "SupportedContentTypes": [],
                    "SupportedCompressionTypes": ["None"],
                    "SupportedInputModes": ["FastFile"],
                },
            ],
        },
    )

Links

Docs: https://boto3.amazonaws.com/v1/documentation/api/latest/reference/services/sagemaker/client/create_algorithm.html

[RyanFitzSimmonsAK]

Hi @fcdl94, thanks for reaching out. FastFile is included in the request syntax because it's a valid option for at least one Sagemaker operation that uses TrainingChannels. However, it is not supported for this operation. This is specified in the documentation for SupportedInputModes,

SupportedInputModes (list) – [REQUIRED]

The allowed input mode, either FILE or PIPE.

In FILE mode, Amazon SageMaker copies the data from the input source onto the local Amazon Elastic Block Store (Amazon EBS) volumes before starting your training algorithm. This is the most commonly used input mode.

In PIPE mode, Amazon SageMaker streams input data from the source directly to your algorithm without using the EBS volume.

Please let me know if you have any follow-up questions.