The TensorFlow container saves the exported models in Ascending order

[chinazm]

We found that there is a bug in sagemaker to export the model:

sagemaker first exported checkpoints as exported models in the output folder and after training there will be a final model zipped and exported. We found sagemaker exported and zipped the models on ascending order not the most recent model exported.

[andremoeller]

Hi @chinazm ,

Thanks for the report! You're right that this isn't the expected behavior -- we'll investigate and comment on this ticket when we have updates.

Thanks again!

[andremoeller]

Hi @chinazm ,

I tried this out and confirmed that we seem to be exporting the wrong checkpoint. I sent this PR to fix this: aws/sagemaker-tensorflow-training-toolkit#58

Thanks!

[andremoeller]

@chinazm

That fix has been merged in. Thanks for the report!