[Serialization] Add is_main_process argument to save_torch_state_dict()
#2648
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
|
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. |
sayakpaul
reviewed
Oct 31, 2024
| @@ -472,3 +472,27 @@ def test_save_torch_state_dict_delete_existing_files( | |||
| assert (tmp_path / "pytorch_model-00001-of-00003.bin").is_file() | |||
| assert (tmp_path / "pytorch_model-00002-of-00003.bin").is_file() | |||
| assert (tmp_path / "pytorch_model-00003-of-00003.bin").is_file() | |||
|
|
|||
|
|
|||
| def test_save_torch_state_dict_not_main_process( | |||
Wauplin
approved these changes
Nov 5, 2024
There was a problem hiding this comment.
Looks good!
Once this is released in huggingface_hub==0.27.0, PRs will be opened to update accelerate's save_model(), transformers' save_pretrained() and diffusers' save_pretrained() to use save_torch_state_dict() directly.
Super exciting!
| except Exception as e: | ||
| logger.warning(f"Error when trying to remove existing '{filename}' from folder: {e}. Continuing...") | ||
| # Only main process should clean up existing files to avoid race conditions in distributed environment | ||
| if is_main_process: |
SunMarc
approved these changes
Nov 5, 2024
|
Thanks for the reviews! |
hanouticelina
deleted the
add-condition-to-cleaning-previous-saved-state-dict
branch
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Outdated suggestions cannot be applied.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 small PR adds
is_main_processparameter tosave_torch_state_dict()to prevent race conditions during distributed environment. This aligns with accelerate's, transformers' and diffusers' implementations and will enable standardization of model saving across these libraries. See #2314 and this internal slack message for more context.Once this is released in huggingface_hub==0.27.0, PRs will be opened to update accelerate's save_model(), transformers' save_pretrained() and diffusers' save_pretrained() to use
save_torch_state_dict()directly.Main changes:
(Following existing implementations in accelerate, transformers and diffusers)
is_main_process=Trueto avoid race conditions during distributed environment.is_main_process=Trueas default parameter.cc @muellerzr, @SunMarc and @sayakpaul for visibility.