Skip empty parameters in gradient reduction #7789
Open
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
Empty parameters (numel=0) cause issues in gradient allreduce when using flatten/unflatten operations. The unflatten operation fails with shape mismatches because empty tensors can't be properly reconstructed from a flattened buffer. This fix skips empty parameters in _get_gradients_for_reduction() since they contribute nothing to gradient reduction anyway. Fixes test_onebit.py::TestOneBitLambEmptyParameters::test Signed-off-by: Masahiro Tanaka <[email protected]>
PKUWZP
approved these changes
Jan 18, 2026
|
|
||
| # Skip empty parameters (numel=0) as they contribute nothing to gradient reduction | ||
| # and cause issues with flatten/unflatten operations | ||
| if param.numel() == 0: |
Collaborator
There was a problem hiding this comment.
@tohtana Very clean fix! The only minor comment is that maybe we can add an explicit test for gradient reduction? It's optional though.
Collaborator
Author
There was a problem hiding this comment.
Thank you for the feedback. Can you elaborate what test you are suggesting? Run only gradient reduction?
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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Outdated suggestions cannot be applied.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.
#7736 fixed an issue with OnebitLamb NaN propagation. With the fix, the optimizer correctly filters out empty parameters, but DeepSpeed engine's gradient allreduce operation (which runs separately from the optimizer) still includes empty parameters' gradients.
This PR addresses the issue by skipping empty parameters (numel=0) in
_get_gradients_for_reduction().