native_layer_norm (for width dim) #3001
Closed
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/3001
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit cec4574 with merge base b1edc3d ( 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: D56005629 |
Summary: We implement `native_layer_norm` which has 3 outputs - normalization of the input tensor according to the given `normalized_shape` - mean - 1/sqrt(var + eps) https://www.internalfb.com/code/fbsource/[8db4b5872791bb88a62ecaa60b667ee4c1b189bf]/fbcode/caffe2/aten/src/ATen/native/native_functions.yaml?lines=3252 According to SS-JIA's suggestion, a model specific implementation is more performant and preferred to a generic one. So we implemented the op in the following optimized way - our current use case has `normalized_shape` of len 1, namely we do the normalization through computing the mean and var at the last width dim - we do the computation in just one shader `native_layer_norm.glsl` without invoking the shaders to compute mean and var respectively - we use [Welford's online algorithm](https://en.wikipedia.org/wiki/Algorithms_for_calculating_variance#Welford's_online_algorithm) to compute mean and variance in one pass Differential Revision: D56005629
copyrightly
force-pushed
the
export-D56005629
branch
from
d0d86da to
cec4574
Compare
|
This pull request was exported from Phabricator. Differential Revision: D56005629 |
copyrightly
added a commit
to copyrightly/executorch
that referenced
this pull request
Apr 12, 2024
Summary: We implement `native_layer_norm` which has 3 outputs - normalization of the input tensor according to the given `normalized_shape` - mean - 1/sqrt(var + eps) https://www.internalfb.com/code/fbsource/[8db4b5872791bb88a62ecaa60b667ee4c1b189bf]/fbcode/caffe2/aten/src/ATen/native/native_functions.yaml?lines=3252 According to SS-JIA's suggestion, a model specific implementation is more performant and preferred to a generic one. So we implemented the op in the following optimized way - our current use case has `normalized_shape` of len 1, namely we do the normalization through computing the mean and var at the last width dim - we do the computation in just one shader `native_layer_norm.glsl` without invoking the shaders to compute mean and var respectively - we use [Welford's online algorithm](https://en.wikipedia.org/wiki/Algorithms_for_calculating_variance#Welford's_online_algorithm) to compute mean and variance in one pass Differential Revision: D56005629
jorgep31415
approved these changes
Apr 15, 2024
|
This pull request has been merged in 74576e8. |
Closed
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:
We implement
native_layer_normwhich has 3 outputsnormalized_shapehttps://www.internalfb.com/code/fbsource/[8db4b5872791bb88a62ecaa60b667ee4c1b189bf]/fbcode/caffe2/aten/src/ATen/native/native_functions.yaml?lines=3252
According to SS-JIA's suggestion, a model specific implementation is more performant and preferred to a generic one. So we implemented the op in the following optimized way
normalized_shapeof len 1, namely we do the normalization through computing the mean and var at the last width dimnative_layer_norm.glslwithout invoking the shaders to compute mean and var respectivelyDifferential Revision: D56005629