[QEff. Finetune]: Added support to sync gradients across devices during backward step only. #477
+64
−31
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
quic-meetkuma
requested review from
quic-rishinr,
ochougul,
quic-hemagnih and
quic-amitraj
as code owners
quic-meetkuma
requested review from
vbaddi,
quic-mamta,
quic-akuruvil and
quic-swatia
Signed-off-by: Meet Patel <meetkuma@qti.qualcomm.com>
Signed-off-by: Meet Patel <meetkuma@qti.qualcomm.com>
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Outdated suggestions cannot be applied.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.
Disabling gradient is necessary when using gradient accumulation for > 1 with ddp enabled.
Currently, we are syncing gradient at every loss.backward() call. When using gradient accumulation, the weight update happens at every n>1 step by calling opt.step(). Only during that step the gradients across each devices should sync with each other.
with model.no_sync() --> context manager solves this issue.
Here, we are not using it but instead setting ddp_model.require_backward_grad_sync to True or False depending on which step we are.