Add pass to extract mutable weights into a .ptd #7798
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
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/7798
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 New FailureAs of commit b67b193 with merge base 948fba6 ( NEW FAILURE - The following job has failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D68121580 |
lucylq
approved these changes
Jan 21, 2025
Summary: Cleaned up the existing pass and fixed a typing error (EP -> PassResult), added another option in backend config to extract only mutable weights (training workflows will do this), fixed the ordering of ET passes and added a warning not to add stuff after memory planning (this pass was actually fine but in general we like having the invariant that memory planning is last), fixed the emitter to prioritize making it external vs mutable. Down the line we will need to support intermixing of mutable and non mutable in the same .ptd (memory regressions not correctness are the stakes), but no one needs that today so deferring. Reviewed By: lucylq Differential Revision: D68121580
JacobSzwejbka
added
the
release notes: api
JacobSzwejbka
force-pushed
the
export-D68121580
branch
from
147b4b2 to
b67b193
Compare
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D68121580 |
YIWENX14
pushed a commit
that referenced
this pull request
Jan 28, 2025
Differential Revision: D68121580 Pull Request resolved: #7798
zonglinpeng
pushed a commit
to zonglinpeng/executorch
that referenced
this pull request
Jan 30, 2025
Differential Revision: D68121580 Pull Request resolved: pytorch#7798
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.
Summary:
Cleaned up the existing pass and fixed a typing error (EP -> PassResult),
added another option in backend config to extract only mutable weights (training workflows will do this),
fixed the ordering of ET passes and added a warning not to add stuff after memory planning (this pass was actually fine but in general we like having the invariant that memory planning is last),
fixed the emitter to prioritize making it external vs mutable. Down the line we will need to support intermixing of mutable and non mutable in the same .ptd (memory regressions not correctness are the stakes), but no one needs that today so deferring.
Reviewed By: lucylq
Differential Revision: D68121580