[BE] replace the extra DeviceMesh _flatten with mesh access #667
Conversation
This file contains 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
… with mesh access" **Summary** pytorch/pytorch#138945 fixes DeviceMesh access on flattened mesh which are constructed from more than 2 meshes. Refer to the fix PR for details if interested. In #592 we avoided this issue by calling `_flatten` instead of direct accessing the flattened mesh. We want to turn back to mesh access which is more straightforward since the fix has been merged in PyTorch. [ghstack-poisoned]
Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #666 **Summary** pytorch/pytorch#138945 fixes DeviceMesh access on flattened mesh which are constructed from more than 2 meshes. Refer to the fix PR for details if interested. In #592 we avoided this issue by calling `_flatten` instead of direct accessing the flattened mesh. We want to turn back to mesh access which is more straightforward since the fix has been merged in PyTorch.
facebook-github-bot
added
the
CLA Signed
XilunWu
changed the title
fegin
approved these changes
Oct 31, 2024
tianyu-l
reviewed
Oct 31, 2024
| # note: this is a workaround of the above logic for old pytorch version | ||
| # where https://github.com/pytorch/pytorch/pull/138945 is not included | ||
| # throw a warning to encourage users to upgrade to a newer pytorch version | ||
| check_if_feature_in_pytorch( |
There was a problem hiding this comment.
@fegin
Since we are already suggesting user to use latest PyTorch nightly, I wonder how necessary it is to include such messages, as multiple features would depend on non-stable release. E.g. for PP, we constantly see new features, without such try-catches.
Not saying this is not good, but there seems to be a trade-off / sacrifice for dev velocity and simplicity.
There was a problem hiding this comment.
I'm okay to remove it. Or we can periodically cleanup (like weekly). In general, this is not just for the end users but also for developers like us who may not up the local PyTorch reguarly.
tianyu-l
reviewed
Oct 31, 2024
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.
Stack from ghstack (oldest at bottom):
Note: This PR is a reland of #666 where the PR was mistakenly merged into a wrong branch.
Summary
pytorch/pytorch#138945 fixes DeviceMesh access on flattened mesh which are constructed from more than 2 meshes. Refer to the fix PR for details if interested.
In #592 we avoided this issue by calling
_flatteninstead of direct accessing the flattened mesh. We want to turn back to mesh access which is more straightforward since the fix has been merged in PyTorch.