distributed with p5 instances

[abdalgader-a]

sagemaker-python-sdk/src/sagemaker/estimator.py

Line 4063 in daa5518

raise ValueError(

Hi, referring to above line - does this still not support SMModelParallel and SMDataParallel when run on p5 instances?

Thanks!

[abdalgader-a]

@ahsan-z-khan any thoughts?

[benieric]

Hi @abdalgader-a, from these docs - https://docs.aws.amazon.com/sagemaker/latest/dg/distributed-data-parallel-support.html#distributed-data-parallel-supported-instance-types

And release notes - https://docs.aws.amazon.com/sagemaker/latest/dg/data-parallel-release-notes.html

It does not seem like it. However, we have recently released a new Training interface called the ModelTrainer which does not contain such check so you should be able to at least boot up a job and verify if the behavior is as you expect - (example)

[abdalgader-a]

@benieric -- thanks for getting back on this. I'll refactor base on the new training interface and test it.