Implementation of powerpropagation as a modifier #1685
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
…tils/helpers.py to still match in the DDP training case, where module. is prepended to layer names
src/sparseml/pytorch/sparsification/pruning/modifier_powerpropagation.py
Show resolved
Hide resolved
abhinavnmagic
previously approved these changes
Oct 4, 2023
anmarques
previously approved these changes
Oct 24, 2023
anmarques
approved these changes
Nov 14, 2023
abhinavnmagic
approved these changes
Nov 16, 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Outdated suggestions cannot be applied.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.
Implement Powerpropagation (raising weights to a power in forward pass), see https://arxiv.org/abs/2110.00296
The functionality is implemented as a Modifier that, when activated, wraps pruned modules (for now, only Conv2d or Linear, but can easily be extended later) in a PowerpropagationWrapper, which does the exponentiation. Once the modifier is deactivated, the architecture change is undone.
Testing: wrote new tests, including for the architecture change and corresponding weight updates. Ran a few epochs on imagenet to make sure that training was looking OK (loss going down/accuracy going up).