Restore generation unwrap state if generate() raises - #7162
Open
YeonwooSung wants to merge 1 commit into
Open
Conversation
Wrap _unwrap_model_for_generation cleanup in try/finally so gradient checkpointing and ZeRO-3 hooks come back after a failed generate.
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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?
_unwrap_model_for_generationdisables gradient checkpointing and, on ZeRO-3, removes DeepSpeed hooks, then restores them only after a successfulyield. Ifgenerate()raises, checkpointing stays off and the hooks stay removed.This wraps the yield in
try/finally, matching_override_model_generation_configin the same file. ZeRO-3add_hooksis also in an innerfinallyso it runs even when generation fails insideGatheredParameters.Adds
test_restores_gradient_checkpointing_on_error.Fixes #6659
Before submitting
AI writing disclosure
We welcome the use of AI tools to help with contributions. For transparency and to help us improve our review process, please indicate the level of AI involvement in this PR.
Who can review?
Anyone in the community is free to review the PR once the tests have passed.
Note
Low Risk
Narrow bugfix to generation context cleanup; behavior on success is unchanged and failures now restore training state instead of leaving the model misconfigured.
Overview
Fixes a cleanup gap in
_unwrap_model_for_generation: whengenerate()(or other code inside the context) raises, gradient checkpointing could stay disabled and, on DeepSpeed ZeRO-3 with gathered params, optimizer hooks might not be reattached.The yield path is now wrapped in
try/finallyso gradient checkpointing is re-enabled on exit, mirroring_override_model_generation_config. For ZeRO-3 withGatheredParameters,add_hooksruns in an innerfinallyso hooks are restored even if generation fails mid-context.Adds
test_restores_gradient_checkpointing_on_errorto assert checkpointing is turned back on after aRuntimeErrorinside the context.Reviewed by Cursor Bugbot for commit 0139109. Bugbot is set up for automated code reviews on this repo. Configure here.