Fix the behavior of collecting 'num_input_tokens_seen' #29099
Merged
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
See huggingface#28791 for more details.
muellerzr
reviewed
Feb 29, 2024
|
Just got it rebased from the main. However something still went wrong in the test, and it seems irrevelant to my PR. |
ArthurZucker
approved these changes
Mar 25, 2024
|
Done, and it has passed all the tests. |
|
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. |
|
thanks for this fix |
hovnatan
pushed a commit
to hovnatan/transformers
that referenced
this pull request
Mar 27, 2024
…9099) fix the behavior of collecting 'num_input_tokens_seen' See huggingface#28791 for more details.
itazap
pushed a commit
that referenced
this pull request
May 14, 2024
fix the behavior of collecting 'num_input_tokens_seen' See #28791 for more details.
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Outdated suggestions cannot be applied.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?
The length of "inputs[main_input_name]" is not guaranteed to be the same when using DDP, which may make the training process hang. Besides, in a distributed setup, it costs a lot to gather the WHOLE input tensors on different workers. It is better to call .numel() first and then .gather().
Fixes #28791
The modified code has already passed the relevant test
tests/trainer/test_trainer_distributed.pywith additional argument 'include_num_input_tokens_seen'.Who can review?
@pacman100
@muellerzr