fix: torch.checkpoint() incorrectly wraps single forward step in original codebase. #274
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
This caused the loss to remain high due to unstable gradients when training with gradient checkpointing enabled. After fixing, accuracy increases as intended when using the gradient checkpoint memory optimization trick.
|
It works. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.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.
Bug Fixes:
Fixes bugs that caused the loss to remain high due to unstable gradients when training with gradient checkpointing enabled. After fixing, the acceptance rate increases as intended when using the gradient checkpoint memory optimization trick.
Modification:
Refactors the draft model’s forward function by separating target model hidden state retrieval and the draft model’s layer flow. Wraps the entire training-time test predictions over the drafting length, removing the torch.checkpoint() loops that previously led to a complicated computation graph and incorrect gradient flows.