[PoC] Enable flexible different layout for same mesh via a util function #1550
+195
−0
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Outdated suggestions cannot be applied.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.
We have seen use cases for more flexible configurations for N-D device mesh especially with MoE and EP. So now we want to implement a util function to support this use case.
For now since we only leverage
_flatten
, so what we can do is to find a way to unflatten to a mesh config which all mesh config can be unflattened into. Then we do flatten on top of that. This might not be very efficient but this is the way we can temporarily unblock users while still figuring out a proper final UX from DeviceMesh side.For now this PR is just for PoC not aim at landing. If we do want to merge this change, I will further polish it to make it mergable into Torchtitan.