Sanity checks for unfinished checkpoints removal

[jbieniusiewi]

What does this PR do ?

The following scenario was recently observed:

• Last checkpoint writing was interrupted due to an irrelevant issue
• Hence last checkpoint was marked as "unfinished" (and there was no other last checkpoint)
• Hence the following training silently skipped the last checkpoint and started from scratch
• As a result, training was not making expected progress despite many training runs

This PR adds some sanity checks to ensure that when the last/end checkpoint is unfinished, it wont get silently skipped and deleted. Instead, user is supposed to remove them manually.

Collection: [Note which collection this PR will affect]

Changelog

• Add specific line by line info of high level changes in this PR.

Usage

• You can potentially add a usage example below

# Add a code snippet demonstrating how to use this 

GitHub Actions CI

The Jenkins CI system has been replaced by GitHub Actions self-hosted runners.

The GitHub Actions CI will run automatically when the "Run CICD" label is added to the PR.
To re-run CI remove and add the label again.
To run CI on an untrusted fork, a NeMo user with write access must first click "Approve and run".

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you add or update any necessary documentation?
• Does the PR affect components that are optional to install? (Ex: Numba, Pynini, Apex etc)
  • Reviewer: Does the PR have correct import guards for all optional libraries?

PR Type:

• New Feature
• Bugfix
• Documentation

If you haven't finished some of the above items you can still open "Draft" PR.

Who can review?

Anyone in the community is free to review the PR once the checks have passed.
Contributor guidelines contains specific people who can review PRs to various areas.

Additional Information

• Related to # (issue)

[maanug-nv]

Thanks @jbieniusiewi . Very simple, and should consistently avoid that training restart we saw, even with auto relaunch.

[maanug-nv]

@jbieniusiewi could you also add the same check for 2.0. The unfinished checkpoint filtering is here in 2.0.

[jbieniusiewi]

Thank you for the review @maanug-nv.
I've added checks for NeMo 2.0. One more approval is needed to merge it, so I re-requested a review.

[maanug-nv]

-@ashors1 Do we have/plan to have tests for AutoResume? Looks like they are in tests/lightning/test_nemo_logger.py. @jbieniusiewi final request, can we include the same test you added for exp manager to NeMoLogger/AutoResume as well?

[jbieniusiewi]

@maanug-nv Sure, I've added tests to tests/lightning/test_nemo_logger.py.

[pzelasko]

Thanks! LGTM