[WIP] Switch gradient checkpointing default to use_reentrant=False (PyTorch recommended)
#43203
+280
−360
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
qgallouedec
changed the title
qgallouedec
changed the title
use_reentrant=False (PyTorch recommended)
|
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. |
qgallouedec
commented
Jan 9, 2026
Comment on lines
-212
to
-227
| @unittest.skip | ||
| def test_training_gradient_checkpointing(self): | ||
| pass | ||
|
|
||
| @unittest.skip( | ||
| reason="This architecture seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124" | ||
| ) | ||
| def test_training_gradient_checkpointing_use_reentrant(self): | ||
| pass | ||
|
|
||
| @unittest.skip( | ||
| reason="This architecture seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124" | ||
| ) | ||
| def test_training_gradient_checkpointing_use_reentrant_false(self): | ||
| pass | ||
|
|
Member
Author
There was a problem hiding this comment.
It seems a big number of this ignored test actually pass. I check them all
Contributor
|
[For maintainers] Suggested jobs to run (before merge) run-slow: align, altclip, aria, autoformer, aya_vision, beit, big_bird, blip, blip_2, canine, chinese_clip, clap, clip, clipseg, colpali, deit |
Contributor
|
View the CircleCI Test Summary for this PR: https://huggingface.co/spaces/transformers-community/circle-ci-viz?pr=43203&sha=435655 |
qgallouedec
changed the title
use_reentrant=False (PyTorch recommended)use_reentrant=False (PyTorch recommended)
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.
Summary
This PR changes our gradient checkpointing default from
use_reentrant=Truetouse_reentrant=False.Two years ago we explicitly set
use_reentrant=Truein #28538 because PyTorch started warning that the default would change in the future, and recommending users choose a value explicitly:At the time, defaulting to
Truewas the safest choice to preserve the behavior of earlier releases.PyTorch now recommends the non-reentrant variant (
use_reentrant=False) see, https://docs.pytorch.org/docs/stable/checkpoint.html, and is moving toward making it the default. Aligning with this upstream recommendation gives us several benefits:Note: training and checkpointing behavior remains functionally equivalent in typical use cases, with the main difference being how activations are recomputed during backward (non-reentrant uses a safer mechanism).