[WIP] ref: decoupled ddp, ddp spawn (finish 3733) #3819
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
ananthsub
reviewed
Oct 3, 2020
ananthsub
reviewed
Oct 3, 2020
ananthsub
reviewed
Oct 3, 2020
| output = self.training_step(args) | ||
| return output | ||
|
|
||
| def barrier(self, name: str = None): |
There was a problem hiding this comment.
Suggested change
| def barrier(self, name: str = None): | |
| def barrier(self, name: Optional[str] = None): |
There was a problem hiding this comment.
good suggestion. i'll change in a separate PR since this affects every accelerator
ananthsub
reviewed
Oct 3, 2020
ananthsub
reviewed
Oct 3, 2020
|
Hello @williamFalcon! Thanks for updating this PR. There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻 Comment last updated at 2020-10-03 17:26:39 UTC |
Co-authored-by: ananthsub <[email protected]>
Codecov Report
@@ Coverage Diff @@
## master #3819 +/- ##
=======================================
+ Coverage 84% 87% +3%
=======================================
Files 111 111
Lines 8783 8947 +164
=======================================
+ Hits 7365 7784 +419
+ Misses 1418 1163 -255 |
2 tasks
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Ok... turned out to fix many problems:
remove a bunch of rank_zero decorators.
We MUST make sure we don't create information asymmetry in children processes. State must be EXACTLY the same. Rank zero is there to not write to disk and that is only where it should be used.
Solved all the issues with DDP about setting VISIBLE_DEVICES and gpus indexing not working. Also solves the issues around calling fit or test or both.
Also solved the issue of how exactly to test ddp since we call other scripts. Overall we should have no regressions going forward because of these tests.
Stability of ddp has been improved dramatically!
Fixes #3422
Fixes #3117
Fixes #3730
Fixes #3605
Fixes #3403
Fixes #2826
Fixes #3071
Fixes #3023
Fixes #2913
Fixes #2590
Fixes #2529
cc @edenafek to add other tickets this fixes
Edit
PR is too big, splitting in parts.
Part 1: #3766
part 2: #3767
part 3: #3770
part 4: #3773
part 5: #3774
part 6: #3783
part 7: #3802
part 8: #3806
TODO:
ie: need
for x in range(5):
trainer.fit()