Repository navigation
Conversation
1 of 4 tasks
nastya236
marked this pull request as ready for review
July 12, 2026 17:55
angeloskath
approved these changes
Jul 20, 2026
angeloskath
left a comment
angeloskath
left a comment
Member
There was a problem hiding this comment.
This looks really good!
I am curious about how you are thinking to approach the re-gathering the weights in the backward. I see a few ways none of which I like.
Either way we should definitely merge this as is. Really nice!
Collaborator
Author
|
Thank you! General high level thought was to checkpoint gather so the forward weights don't stay in the graph, but you probably though about it more than me.. |
3 of 4 tasks
1 task done
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Fully sharded data parallel.
It is incomplete, because bf16 weights that are gathered during forward are not resharded and used in backward. So we can cut another ~6GB by resharding and gathering back during backward.
Peak memory difference between sharded optimizer and this PR is 14 GB [4B * 4 = 16GB -- master copy weights, therefore each rank holds only 16GB / 8 =2 GB 2 instead of 16GB] as expected.