Remove layer norm from the default quantizer, add one that has it #10182
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
Summary: Layer norm is not performing great in quantized mode, and is currently using a split scheme (weights are quantized, activations are not). In most cases, it's actually much faster to keep it fp32, so this diff removes it from the default quantizer. We add a CadenceWithLayerNormQuantizer for easy access to the current behavior, which can be good in some cases (mostly if quantizing layer norm will help extend the quantized liveness). Differential Revision: D72941790
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/10182
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 817c459 with merge base 82ff404 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
|
This pull request was exported from Phabricator. Differential Revision: D72941790 |
zonglinpeng
approved these changes
Apr 15, 2025
keyprocedure
pushed a commit
to keyprocedure/executorch
that referenced
this pull request
Apr 21, 2025
Differential Revision: D72941790 Pull Request resolved: pytorch#10182
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Outdated suggestions cannot be applied.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.
Summary:
Layer norm is not performing great in quantized mode, and is currently using a split scheme (weights are quantized, activations are not). In most cases, it's actually much faster to keep it fp32, so this diff removes it from the default quantizer.
We add a CadenceWithLayerNormQuantizer for easy access to the current behavior, which can be good in some cases (mostly if quantizing layer norm will help extend the quantized liveness).
Differential Revision: D72941790