In pipeline parallelism: Use same dtype for receive and send tensor when initializing p2p communication.#165539
Closed
BlueCrescent wants to merge 2 commits intopytorch:mainfrom
Conversation
… when initializing p2p communication.
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/165539
Note: Links to docs will display an error until the docs builds have been completed. ⏳ 1 Pending, 1 Unrelated FailureAs of commit 5974b80 with merge base 5c583e2 ( FLAKY - The following job failed but was likely due to flakiness present on trunk:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
pytorch-bot
bot
added
the
oncall: distributed
H-Huang
added
release notes: distributed (pipeline)
H-Huang
approved these changes
Oct 15, 2025
Member
H-Huang
left a comment
H-Huang
left a comment
There was a problem hiding this comment.
Thank you for root causing and fixing. Send/recv with mismatched dtypes indeed causes undefined behavior so we didn't see this in our runs. Really appreciate your investigation!
Member
|
@pytorchbot merge |
Collaborator
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Chao1Han
pushed a commit
to Chao1Han/pytorch
that referenced
this pull request
Oct 21, 2025
…hen initializing p2p communication. (pytorch#165539) When initializing the p2p communication for pipeline parallelism, currently different default dtypes are used for the send and receive tensor here: https://github.com/pytorch/pytorch/blob/5c583e2573f29243742e00b9fa36b266c5c78bb3/torch/distributed/pipelining/stage.py#L935-L936 This caused hard to trace issues when training on multiple nodes. Multiple stages on one node seem to work for some reason which probably caused the unit tests not to catch this. Fixes pytorch#165143 Pull Request resolved: pytorch#165539 Approved by: https://github.com/H-Huang
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
4 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.
When initializing the p2p communication for pipeline parallelism, currently different default dtypes are used for the send and receive tensor here:
pytorch/torch/distributed/pipelining/stage.py
Lines 935 to 936 in 5c583e2
This caused hard to trace issues when training on multiple nodes. Multiple stages on one node seem to work for some reason which probably caused the unit tests not to catch this.
Fixes #165143
cc @H-Huang @awgu @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k @pragupta @msaroufim @dcci