Simplified setup of optimizers in FSDP #17309
Merged
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
awaelchli
added
feature
for more information, see https://pre-commit.ci
…orch' into feature/fsdp-use-orig-params-pytorch
awaelchli
requested review from
Borda,
carmocca,
edenlightning,
justusschock,
lantiga and
williamFalcon
as code owners
Contributor
Borda
approved these changes
Apr 11, 2023
carmocca
approved these changes
Apr 11, 2023
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 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Outdated suggestions cannot be applied.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.
What does this PR do?
Resolves a limitation in FSDP for users of PyTorch 2.0+. Previously, to use FSDP you had to make code modifications. In particular, you had to write
optimizer = Adam(self.trainer.model.parameters)instead of justoptimizer = Adam(self.parameters). This can be avoided by enablinguse_orig_paramsin the FSDP wrapper. This PR enables this for users of PyTorch >= 2.0 automatically. This also enables multiple parameter groups which was previously not possible.This PR goes alongside the equivalent change in Fabric: #17305
cc @Borda @awaelchli @carmocca