Fix gradient accumulation for ShardedDataParallel

[ananthsub]

What does this PR do?

Followup to this comment: #9101 (comment)

ShardedDataParallel supports the no_sync context manager too: https://fairscale.readthedocs.io/en/latest/_modules/fairscale/nn/data_parallel/sharded_ddp.html#ShardedDataParallel.no_sync

But we're not taking advantage of it now due to this check:https://github.com/PyTorchLightning/pytorch-lightning/blob/9d62f248476c6358d8707188f7b20fafa79f8a4f/pytorch_lightning/plugins/training_type/parallel.py#L131-L133

as the model here is wrapped with ShardedDataParallel not DistributedDataParallel

Breaking the inheritance chain in these plugins will make these opportunities clearer.

@SeanNaren n00b question: do you have suggestions for how to verify this a unit/integration test? especially to prevent future regressions

Does your PR introduce any breaking changes? If yes, please list them.

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 list all the breaking changes introduced by this pull request?
• Did you update the CHANGELOG? (not for typos, docs, test updates, or internal minor changes/refactorings)

PR review

Anyone in the community is welcome to review the PR.
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?

Make sure you had fun coding 🙃

[codecov]

Codecov Report

Merging #9122 (64e3e7c) into master (d022f6f) will decrease coverage by 4%.
The diff coverage is 50%.

❗ Current head 64e3e7c differs from pull request most recent head bc9b7ee. Consider uploading reports for the commit bc9b7ee to get more accurate results

@@           Coverage Diff           @@
##           master   #9122    +/-   ##
=======================================
- Coverage      93%     89%    -4%     
=======================================
  Files         179     179            
  Lines       15303   15317    +14     
=======================================
- Hits        14200   13590   -610     
- Misses       1103    1727   +624     

[tchaton]

LGTM !

[SeanNaren]

Thanks for the PR @ananthsub!

What we can ensure in a test is that the context manager is called correctly during grad accumulation, maybe by mocking/patching the function?

[ananthsub]

@tchaton I'll add a test first before merging. still on my todo