fix accelerator prepare during eval only mode #24014
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
|
The documentation is not available anymore as the PR was closed or merged. |
sgugger
reviewed
Jun 5, 2023
Collaborator
sgugger
left a comment
sgugger
left a comment
There was a problem hiding this comment.
Unless I'm missing something, this changes the whole logic of the evaluation in the Trainer and should not be done.
src/transformers/trainer.py
Outdated
| model = self._wrap_model(self.model, training=False, dataloader=dataloader) | ||
|
|
||
| if len(self.accelerator._models) == 0 and model is self.model: | ||
| model = self.accelerator.prepare(model) |
Collaborator
There was a problem hiding this comment.
No we only want to do this for DeepSpeed, not all the time. Putting a model in DistributedDataParallel just for evaluation will waste some memory.
Contributor
Author
There was a problem hiding this comment.
I do agree on the DDP case and hence I didn't update it earlier but as mentioned below we will be missing mixed precision coverage for eval-only mode
Contributor
Author
|
The thing is that mixed precision application for eval only mode won't work unless we prepare model |
sgugger
approved these changes
Jun 7, 2023
novice03
pushed a commit
to novice03/transformers
that referenced
this pull request
Jun 23, 2023
* fix mixed precision prep during eval only mode * update to address comments * update to reflect the changes in accelerate
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Outdated suggestions cannot be applied.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?
preparemethod is happening only during training loop. If the user is directly doingeval/predictwithout the training loop, the model isn't prepared leading to wrong behaviour. This PR is aimed at fixing it.evaluation_modeaccelerate#1540