Save tensors in context of memory_efficient_linear #3413
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
tohtana
requested review from
jeffra,
tjruwase,
samyam and
mrwyattii
as code owners
tjruwase
approved these changes
May 1, 2023
|
@tjruwase Thank you for merging this PR! As I mentioned, I didn't understand the intention of saving tensors in a global map. |
|
@tohtana, I think your solution is the correct one. |
|
Thank you for your reviewing, @tjruwase! |
This was referenced May 5, 2023
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Outdated suggestions cannot be applied.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.
By default,
torch.nn.functional.linearis replaced withLinearFunctionForZeroStage3. However,LinearFunctionForZeroStage3causes memory leak in some usecases.In PEFT's LoRA mentioned in #3002, the weight is passed after 'transpose'.
LinearFunctionForZeroStage3 saves the weight in a map and the key is the object's ID. But a new transposed weight is created and the ID changes for each iteration. So the saved weights will increase through iterations.
This PR simply saves weight and bias in the context instead of the IDs.
I don't understand the intention of using IDs to store the weight. If saving IDs instead of tensors is a crucial part of this module, we need another approach to fix this.