Introduce all_reduce_hook to support gradient aggregation across replica groups. #7764
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
…ica groups. Signed-off-by: zhengchenyu <[email protected]>
zhengchenyu
requested review from
loadams,
tjruwase and
tohtana
as code owners
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.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.
Using replica groups offers the following advantages:
For stage 3, it ensures that parameter gather during forward and backward occurs only within the replica group.
Checkpointing is performed only on
replica_group_rank=0, guaranteeing constant checkpoint world size and avoiding the universal checkpoint transformations during scaling up or down.We can achieve gradient all reduce within the replica group after backward and before optimizer.step, but we must wait for all buckets to complete, thus can not leverage concurrency advantages.
I know MICS has similar functionality, but currently only supports zero stage 3. Additionally, I want to use this feature for compatibility with architectures like TorchFT.