Fix lr scheduler state not being dumped to checkpoint in deepspeed strategy

[awaelchli]

What does this PR do?

Part of #11188

The added assertion fails on master.

Before submitting

• Was this discussed/approved via a GitHub issue? (not for typos and docs)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure your PR does only one thing, instead of bundling different changes together?
• Did you make sure to update the documentation with your changes? (if necessary)
• Did you write any new necessary tests? (not for typos and docs)
• Did you verify new and existing tests pass locally with your changes?
• Did you update the CHANGELOG? (not for typos, docs, test updates, or internal minor changes/refactorings)

PR review

Anyone in the community is free to review the PR once the tests have passed.
Before you start reviewing make sure you have read Review guidelines. In short, see the following bullet-list:

• Is this pull request ready for review? (if not, please submit in draft mode)
• Check that all items from Before submitting are resolved
• Make sure the title is self-explanatory and the description concisely explains the PR
• Add labels and milestones (and optionally projects) to the PR so it can be classified

Did you have fun?

I made sure I had fun coding 🙃

Part of #1 (it's a lie, this is just here to avoid noisy GitHub bot)

cc @Borda @SeanNaren @awaelchli @rohitgr7

[SeanNaren]

Thanks for fixing this @awaelchli. I assume based on the logic, the scheduler will be loaded if found in the state dict elsewhere in the code

[tridao]

Thanks for fixing this @awaelchli. I assume based on the logic, the scheduler will be loaded if found in the state dict elsewhere in the code

I'm not sure that's the case, since the Deepspeed plugin has lightning_restore_optimizer_and_schedulers set to False.
Would the checkpoint_connector ignore the lr_schedulers key when restoring?

def restore_optimizers_and_schedulers(self) -> None:
        """Restores the optimizers and learning rate scheduler states from the pre-loaded checkpoint."""
        if not self._loaded_checkpoint or not self.trainer.strategy.lightning_restore_optimizer_and_schedulers:
            return

[awaelchli]

Yes you are right, this needs to be taken care of too.

[eelxpeng]

Right. I tried with current commits. Though lr_schedulers are saved, the lr_scheduler state will not be restored.

[codecov]

Codecov Report

Merging #11307 (395d62a) into master (7eab379) will decrease coverage by 4%.
The diff coverage is 0%.

@@           Coverage Diff            @@
##           master   #11307    +/-   ##
========================================
- Coverage      92%      88%    -4%     
========================================
  Files         194      194            
  Lines       16965    16965            
========================================
- Hits        15593    14960   -633     
- Misses       1372     2005   +633