[FSDP][optim_state_dict] Make the new optimizer allgather fusion work with fine-tuning models #110540
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
… with fine-tuning models With use_orig_params=True, it is possible that some parameters in a FlatParameter are in the optimizer while others are frozen. This PR makes the allgather fusion logic support the case. Differential Revision: [D49922028](https://our.internmc.facebook.com/intern/diff/D49922028/) [ghstack-poisoned]
fegin
requested review from
mrshenli,
zhaojuanmao,
rohan-varma,
H-Huang,
awgu,
kwen2501,
wanchaol,
fduwjj,
wz337,
kiukchung and
d4l3k
as code owners
pytorch-bot
bot
added
the
release notes: distributed (fsdp)
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/110540
Note: Links to docs will display an error until the docs builds have been completed. ✅ You can merge normally! (3 Unrelated Failures)As of commit 9135b1e with merge base 4069d1d ( FLAKY - The following job failed but was likely due to flakiness present on trunk:
UNSTABLE - The following jobs failed but were likely due to flakiness present on trunk and has been marked as unstable:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
fegin
added a commit
that referenced
this pull request
Oct 4, 2023
… with fine-tuning models With use_orig_params=True, it is possible that some parameters in a FlatParameter are in the optimizer while others are frozen. This PR makes the allgather fusion logic support the case. Differential Revision: [D49922028](https://our.internmc.facebook.com/intern/diff/D49922028/) ghstack-source-id: 202913384 Pull Request resolved: #110540
…fusion work with fine-tuning models" With use_orig_params=True, it is possible that some parameters in a FlatParameter are in the optimizer while others are frozen. This PR makes the allgather fusion logic support the case. Differential Revision: [D49922028](https://our.internmc.facebook.com/intern/diff/D49922028/) [ghstack-poisoned]
fegin
added a commit
that referenced
this pull request
Oct 4, 2023
… with fine-tuning models Pull Request resolved: #110540 With use_orig_params=True, it is possible that some parameters in a FlatParameter are in the optimizer while others are frozen. This PR makes the allgather fusion logic support the case. ghstack-source-id: 202914577 @exported-using-ghexport Differential Revision: [D49922028](https://our.internmc.facebook.com/intern/diff/D49922028/)
…fusion work with fine-tuning models" With use_orig_params=True, it is possible that some parameters in a FlatParameter are in the optimizer while others are frozen. This PR makes the allgather fusion logic support the case. Differential Revision: [D49922028](https://our.internmc.facebook.com/intern/diff/D49922028/) [ghstack-poisoned]
fegin
added a commit
that referenced
this pull request
Oct 4, 2023
… with fine-tuning models Pull Request resolved: #110540 With use_orig_params=True, it is possible that some parameters in a FlatParameter are in the optimizer while others are frozen. This PR makes the allgather fusion logic support the case. ghstack-source-id: 202915043 @exported-using-ghexport Differential Revision: [D49922028](https://our.internmc.facebook.com/intern/diff/D49922028/)
fegin
added
ciflow/trunk
…fusion work with fine-tuning models" With use_orig_params=True, it is possible that some parameters with the same FlatParameter are in the optimizer while others parameters are frozen. This PR makes the allgather fusion logic support the case. Differential Revision: [D49922028](https://our.internmc.facebook.com/intern/diff/D49922028/) [ghstack-poisoned]
fegin
added a commit
that referenced
this pull request
Oct 4, 2023
… with fine-tuning models Pull Request resolved: #110540 With use_orig_params=True, it is possible that some parameters with the same FlatParameter are in the optimizer while others parameters are frozen. This PR makes the allgather fusion logic support the case. ghstack-source-id: 202917259 @exported-using-ghexport Differential Revision: [D49922028](https://our.internmc.facebook.com/intern/diff/D49922028/)
awgu
approved these changes
Oct 4, 2023
| # 1.) the rank does not own any part of the original parameter. | ||
| # As a result, there is no corresponding optimizer state on | ||
| # the rank as well. | ||
| # 2.) the parameter is frozen and no optimizer state for the |
There was a problem hiding this comment.
This case can happen when the parameter changes between frozen and non-frozen, so it is currently frozen but still has optimizer state?
rohan-varma
approved these changes
Oct 4, 2023
There was a problem hiding this comment.
Similar question to @awgu - what happens when a parameter transitions state from frozen -> nonfrozen, or vice-versa during training?
…fusion work with fine-tuning models" With use_orig_params=True, it is possible that some parameters with the same FlatParameter are in the optimizer while others parameters are frozen. This PR makes the allgather fusion logic support the case. Differential Revision: [D49922028](https://our.internmc.facebook.com/intern/diff/D49922028/) [ghstack-poisoned]
fegin
added a commit
that referenced
this pull request
Oct 4, 2023
… with fine-tuning models Pull Request resolved: #110540 With use_orig_params=True, it is possible that some parameters with the same FlatParameter are in the optimizer while others parameters are frozen. This PR makes the allgather fusion logic support the case. ghstack-source-id: 202943644 @exported-using-ghexport Differential Revision: [D49922028](https://our.internmc.facebook.com/intern/diff/D49922028/)
|
The latest version supports the no_grad -> grad -> no_grad transition and changes the UT to verify the usage. |
…fusion work with fine-tuning models" With use_orig_params=True, it is possible that some parameters with the same FlatParameter are in the optimizer while others parameters are frozen. This PR makes the allgather fusion logic support the case. Differential Revision: [D49922028](https://our.internmc.facebook.com/intern/diff/D49922028/) [ghstack-poisoned]
fegin
added a commit
that referenced
this pull request
Oct 4, 2023
… with fine-tuning models Pull Request resolved: #110540 With use_orig_params=True, it is possible that some parameters with the same FlatParameter are in the optimizer while others parameters are frozen. This PR makes the allgather fusion logic support the case. ghstack-source-id: 202944435 @exported-using-ghexport Differential Revision: [D49922028](https://our.internmc.facebook.com/intern/diff/D49922028/)
…fusion work with fine-tuning models" With use_orig_params=True, it is possible that some parameters with the same FlatParameter are in the optimizer while others parameters are frozen. This PR makes the allgather fusion logic support the case. Differential Revision: [D49922028](https://our.internmc.facebook.com/intern/diff/D49922028/) [ghstack-poisoned]
fegin
added a commit
that referenced
this pull request
Oct 4, 2023
… with fine-tuning models Pull Request resolved: #110540 With use_orig_params=True, it is possible that some parameters with the same FlatParameter are in the optimizer while others parameters are frozen. This PR makes the allgather fusion logic support the case. ghstack-source-id: 202957449 @exported-using-ghexport Differential Revision: [D49922028](https://our.internmc.facebook.com/intern/diff/D49922028/)
|
@pytorchbot merge (Initiating merge automatically since Phabricator Diff has merged) |
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
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.
Stack from ghstack (oldest at bottom):
With use_orig_params=True, it is possible that some parameters with the same FlatParameter are in the optimizer while others parameters are frozen. This PR makes the allgather fusion logic support the case.
Differential Revision: D49922028