Reenable the distributed checkpointing test#8424
Merged
Conversation
tengyifei
approved these changes
Dec 2, 2024
rpsilva-aws
pushed a commit
to rpsilva-aws/xla
that referenced
this pull request
Dec 6, 2024
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
2 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
This is follow up of #8386.
In the previous pr I found that someone during fallback the pytorch will try to update an existing XLATensor with a CPU tesnor with different shape. In that case we need to remove the sharding spec otherwise there will be a shape mismatch. However I found that in the distributed point we will swap the existing XLATensor with the cpu tensor and it seems like we want to keep the sharding spec.
@jonb377 one concern I have is that test only test the single host, I felt like if it is a actual multi-host case the CPU tensor withh have different shape(sharded) than the shardingspec? I am not sure if we have such test somewhere. Even if we clear the shardingspec after a
torch_xla.sync()the tensor will be moved to the device, but most likely replicated. I am a bit worried if I am breaking the distributed checkpointing here.