Bunch of FSDP improvements#3671
Merged
Merged
Conversation
S1ro1
commented
Jul 7, 2025
| if self.backend == "tccl": | ||
| local_rank = os.environ.get("LOCAL_RANK", -1) | ||
| torch.sdaa.set_device(f"sdaa:{local_rank}") | ||
| if ( |
Contributor
Author
There was a problem hiding this comment.
This had to be moved, as it was in a deepspeed branch before, which was wrong
S1ro1
commented
Jul 7, 2025
| if fsdp_plugin.state_dict_type == StateDictType.FULL_STATE_DICT: | ||
| if type(model) is not FSDP and accelerator.process_index != 0: | ||
| if ( | ||
| type(model) is not FSDP and not isinstance(model, torch.distributed.fsdp.FSDPModule) |
Contributor
Author
There was a problem hiding this comment.
In here, we actually early exit if FSDP1, which shouldn't be the case in FSDP2 as the set_model_state_dict timeouts
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
SunMarc
approved these changes
Jul 9, 2025
SunMarc
left a comment
Member
There was a problem hiding this comment.
Nice ! Thanks for the improvements !
S1ro1
added a commit
that referenced
this pull request
Jul 9, 2025
* Feat: split tests * Feat: finito * Fix * Final, tests pass
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
3 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
First I intended just to make the tests better, but I found some issues with the underlying code so I'm fixing that here too (can split if needed), most of it related to state loading, which somehow didn't get caught in the CI
sync_module_stateswhich we used in FSDP1 we just exit on non-main ranks.