Add unit test to ensure no gradients sync when calling ddp.module(input) #20282
Conversation
This file contains hidden or 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
pytorchbot
added
the
oncall: distributed
chenyangyu1988
force-pushed
the
export-D15263155
branch
from
ccc446f to
e3cd089
Compare
mrshenli
approved these changes
May 8, 2019
Contributor
mrshenli
left a comment
mrshenli
left a comment
There was a problem hiding this comment.
Thanks for adding the test. I feel this test should only temporarily live to make sure changes in PyTorch does not break your use cases. And it should retire when we have a better solution (e.g., a context manager as @apaszke suggested).
test/test_c10d.py
Outdated
Contributor
There was a problem hiding this comment.
Let's do 4 iterations to make sure that things are still correct after the first sync.
test/test_c10d.py
Outdated
Contributor
There was a problem hiding this comment.
Shall we add a comment to say that this would only work when there is a single model replica per process. Otherwise, _sync_params would broadcast params from replica 0 to others and erase accumulated grads.
chenyangyu1988
force-pushed
the
export-D15263155
branch
from
e3cd089 to
2132e23
Compare
…ut) (pytorch#20282) Summary: Pull Request resolved: pytorch#20282 Add unit test to ensure no gradients sync when calling ddp.module(input), e.g not invoking prepare_for_backward PyText is depending on DDP for data parallel distributed training. To support accumulate gradients locally before gradients sync, we are calling orig_model.forward instead of ddp_model.forward. Add a unit test to avoid changes break the assumption. Reviewed By: mrshenli Differential Revision: D15263155 fbshipit-source-id: f5bd884616b2783064f383820dc6397f8608c3c6
chenyangyu1988
force-pushed
the
export-D15263155
branch
from
2132e23 to
2f194b0
Compare
Contributor
|
This pull request has been merged in 2019f6c. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
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.
Summary:
Add unit test to ensure no gradients sync when calling ddp.module(input), e.g not invoking prepare_for_backward
PyText is depending on DDP for data parallel distributed training. To support accumulate gradients locally before gradients sync, we are calling orig_model.forward instead of ddp_model.forward. Add a unit test to avoid changes break the assumption.
Differential Revision: D15263155