Factorized linear supports implementation switch and gradient checkpoint #26
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
There was a problem hiding this comment.
Thanks @JeremieMelo, this looks great, just made a few suggestions on the API.
| n_layers : int, default is 1 | ||
| number of linear layers to be parametrized with a single factorized tensor | ||
| bias : bool, default is True | ||
| with_cp : bool |
There was a problem hiding this comment.
I think we can just call it checkpointing to be more explicit?
There was a problem hiding this comment.
Updated in the newest commit.
| device : PyTorch device to use, default is None | ||
| dtype : PyTorch dtype, default is None | ||
| """ | ||
| def __init__(self, in_tensorized_features, out_tensorized_features, bias=True, | ||
| factorization='cp', rank='same', n_layers=1, device=None, dtype=None): | ||
| factorization='cp', rank='same', implementation='factorized', n_layers=1, | ||
| with_cp=False, device=None, dtype=None): |
There was a problem hiding this comment.
Instead of with_cp, we can use checkpointing
There was a problem hiding this comment.
updated in the newest commit.
Sure, could you create a suggested change to replace all 'with_cp' to 'checkpointing', so I can directly add it to the batched commit. Thanks. Co-authored-by: Jean Kossaifi <jean.kossaifi@gmail.com>
with_cp to checkpointing, move weight out of _inner_forward
|
Looks good, thanks @JeremieMelo, merging! |
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Outdated suggestions cannot be applied.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.
support implementation switches between factorized and reconstructed
gradient checkpointing for memory-efficient training-mode forward function.