[fix] handle FSDP DTensor in broadcast_from_megatron_pp#113
Open
yxs wants to merge 1 commit intoISEEKYAN:mainfrom
Open
[fix] handle FSDP DTensor in broadcast_from_megatron_pp#113yxs wants to merge 1 commit intoISEEKYAN:mainfrom
yxs wants to merge 1 commit intoISEEKYAN:mainfrom
Conversation
Megatron FSDP (ZeRO-3) stores parameters as DTensors. When export_weights broadcasts params across PP ranks, torch.distributed.broadcast() triggers DTensor dispatch and fails because the PP group is not in the DTensor's DeviceMesh. Fix: call DTensor.full_tensor() to materialize the full parameter before broadcasting.
Author
|
@ISEEKYAN could you please take a look. |
4 tasks
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What does this PR do?
Megatron FSDP (ZeRO-3) stores parameters as DTensors. When
export_weightsbroadcasts params across PP ranks,torch.distributed.broadcast()triggers DTensor dispatch and fails:Fix: call
DTensor.full_tensor()to materialize the full parameter before broadcasting. Backward compatible, no-op for non-FSDP parameters.