update logic to inject FastForwardSampler / CaptureIterableDataset 2/n

[tchaton]

What does this PR do?

Should be merged after #8364

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

• Could cause different behaviour depending on the seed of samplers within the iterable dataset.

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)
• Did you list all the breaking changes introduced by this pull request?

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?

Make sure you had fun coding 🙃

[codecov]

Codecov Report

Merging #8366 (36076ed) into master (850416f) will increase coverage by 4%.
The diff coverage is 100%.

@@           Coverage Diff           @@
##           master   #8366    +/-   ##
=======================================
+ Coverage      88%     93%    +4%     
=======================================
  Files         168     167     -1     
  Lines       13985   13971    -14     
=======================================
+ Hits        12355   12940   +585     
+ Misses       1630    1031   -599     

[yifuwang]

Hi @tchaton, should we have replaced the sampler even when self.accelerator_connector.replace_sampler_ddp is False? I'm not familiar with the logic here, but a few of our tests failed because replace_sampler is now invoked after this change. Specifically, the test uses a custom sampler, but _resolve_batch_sampler expects to be able to re-construct the sampler following way:

if (batch_sampler is not None and type(batch_sampler) is not BatchSampler) or is_predicting:
    batch_sampler = type(batch_sampler)(
        sampler,
        batch_size=batch_sampler.batch_size,
        drop_last=(False if is_predicting else batch_sampler.drop_last),
    )

However, the custom sampler in question doesn't have the expected fields.

[tchaton]

It won't actually replace it if replace_sampler_ddp. But the code should be refactored as it does more than just replace the sampler.

[yifuwang]

Is it fine if we only invoke replace_sampler in the if-clause like before? I've confirmed that it would "fix" our test.