SageMaker training with FSx: what is "directory_path"

Question

SageMaker can train on FSx data. One [SageMaker SDK parameter for FSx training][1] is `directory_path`. Where do we find that?

[1]: https://sagemaker.readthedocs.io/en/stable/overview.html

Accepted Answer

FSx for Lustre is a file system that you can use to provide high performance for ML training workloads. The directory_path should point to the location on your file system where your dataset is stored.

In the example in the docs:
directory_path='/fsx/tensorflow',

/fsx is the directory you define on your compute instances where you are mounting the file system
/tensorflow would represent a folder within the fsx directory

If you are using an S3-linked FSx for Lustre file system /tensorflow would be a prefix within your S3-linked bucket.

SageMaker training with FSx: what is "directory_path"

Contenido relevante