[Fix] Move log value to cpu.

[tchaton]

What does this PR do?

This PR introduces fixes for gpu leak.
It adds a move_metrics_to_cpu parameter to the Trainer to force all result objects to be moved to cpu.

Fixes #4556

Before submitting

• Was this discussed/approved via a Github issue? (no need for typos and docs improvements)
• 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? Otherwise, we ask you to create a separate PR for every change.
• Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?
• Did you verify new and existing tests pass locally with your changes?
• If you made a notable change (that affects users), did you update the CHANGELOG?

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 in short, see 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; Bugfixes should be including in bug-fix release milestones (m.f.X) and features should be included in (m.X.b) releases.

Did you have fun?

Make sure you had fun coding 🙃

[Vozf]

Shouldn't there also be a detach()?

[tchaton]

        # no metrics should be logged with graphs
        if not enable_graph and isinstance(value, torch.Tensor):
            value = value.detach()

        # sync across workers when using distributed training
        sync_fn = sync_fn or sync_ddp_if_available
        if sync_dist and isinstance(value, (torch.Tensor, numbers.Number)):
            value = sync_fn(value, group=sync_dist_group, reduce_op=sync_dist_op)

        # no need to keep on gpu
        if isinstance(value, torch.Tensor) and value.is_cuda:
            value = value.cpu()

detach is called just before.

[Vozf]

Have you also checked the same thing with validation loop?

[tchaton]

Hey @Vozf,

Did you join Slack ? If no, please do. So we can quickly resolve this bug :)

Best,
T.C

[Vozf]

Joined now as Vozf. I don't have any more questions except if it works for validation and testing if it fixes the bug in general :)

[codecov]

Codecov Report

Merging #4592 (5be63ce) into master (7e08b0d) will decrease coverage by 0%.
The diff coverage is 75%.

@@          Coverage Diff           @@
##           master   #4592   +/-   ##
======================================
- Coverage      93%     93%   -0%     
======================================
  Files         116     116           
  Lines        8883    8907   +24     
======================================
+ Hits         8278    8291   +13     
- Misses        605     616   +11     

[s-rog]

Do we need a test for this as well? (assert tensor device cpu)

[Borda]

lets try the .detach() intead?

[justusschock]

This is fine to me, just some minor question

[SeanNaren]

@edenlightning swapping this to 1.1 since its tied to the logging refactor.