Support DTensor params in local_sgd/diloco #168
Merged
Conversation
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
facebook-github-bot
added
the
CLA Signed
H-Huang
force-pushed
the
diloco_titan
branch
2 times, most recently
from
9a89a69 to
6ba0eee
Compare
H-Huang
commented
Apr 18, 2025
d4l3k
approved these changes
Apr 18, 2025
|
Without the Dtensor conditionals, the operations like |
fegin
reviewed
Apr 21, 2025
|
|
||
| def _restore_parameters(self) -> None: | ||
| with torch.no_grad(): | ||
| # TODO: consider running copy on a separate stream | ||
| for name, p in self._model.named_parameters(): | ||
| p.data.copy_(self.original_parameters[name], non_blocking=False) | ||
| if isinstance(p, DTensor): |
There was a problem hiding this comment.
If p is a DTensor, does p.copy_(), instead of p.data.copy_(), work?
There was a problem hiding this comment.
IIRC it didn't work and causes a segfault
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Outdated suggestions cannot be applied.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.
Tested in torchtitan (pytorch/torchtitan#1122) when the model is sharded with FSDP we need to convert the params to local tensors, update, then back to DTensors.
Ideally I would like to add a test in the integration tests but that would require us to set up the device_mesh / FSDP for a model