Add Pytorch Native AMP support in Trainer #6151
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
sgugger
approved these changes
Jul 30, 2020
julien-c
approved these changes
Jul 30, 2020
LysandreJik
approved these changes
Jul 31, 2020
Mehrad0711
pushed a commit
to Mehrad0711/transformers
that referenced
this pull request
Aug 3, 2020
* fixed type; add Pytorch Native CUDA AMP support * reverted commit on modeling_utils * confirming to HF black formatting rule * changed bool value of _use_apex * scaler support for gradient clipping * fix inplace operation of clip_grad_norm * removed not while version comparison
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Outdated suggestions cannot be applied.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.
Pytorch 1.6 introduces Native AMP support. This eliminates the need to build and install Apex and provides improvements over problems highlighted in Apex #818 and flexibility. This is the recommended way to use AMP.
With this PR, Trainer will automatically use Pytorch's native AMP if 1.6 version is installed, otherwise, it will use Apex.
This PR will close #6115.