fix ram efficient fsdp init #25686
Merged
fix ram efficient fsdp init #25686
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 documentation is not available anymore as the PR was closed or merged. |
4 tasks
parambharat
pushed a commit
to parambharat/transformers
that referenced
this pull request
Sep 26, 2023
blbadger
pushed a commit
to blbadger/transformers
that referenced
this pull request
Nov 8, 2023
EduardoPach
pushed a commit
to EduardoPach/transformers
that referenced
this pull request
Nov 18, 2023
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Outdated suggestions cannot be applied.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.
What does this PR do?
TrainingArgumentsobject, torch distributed process group won't be initialized and as such when FSDP is enabled via accelerate config, it will end up initializing the model with random weights on all ranks as theis_fsdp_enabled_and_dist_rank_0function will always returnFalse. This results in NaN losses. Quite a journey to uncover this bug. This PR fixes it.