share inflight registry between PartitionedParameterCoordinators #3462
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
HeyangQin
requested review from
jeffra,
tjruwase,
samyam and
mrwyattii
as code owners
tjruwase
approved these changes
May 12, 2023
|
I verified this works great together with our patch on the Lightning side (ready to be merged soon) 🎉 |
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Outdated suggestions cannot be applied.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.
This is a collaborative effort with the Lightning team to solve #3068 and #3156. More discussion at Lightning-AI/pytorch-lightning#17523
There could be multiple PartitionedParameterCoordinator instances, yet they currently manage the parameters in a standalone manner. Let's say we have PartitionedParameterCoordinator A and B. When A puts some parameters inflight, B is not aware of that and when B tries to use these parameters it will just error out. This PR addresses this issue by making the __InflightParamRegistry shared among all PartitionedParameterCoordinator instances. Different from the #3380, this PR would bind the registry to the model so it doesn't break multi-model training