Add MLFloat16 support for LayerNormalization, SkipLayerNormalization#22063
Merged
amarin16 merged 20 commits intomicrosoft:mainfrom Sep 24, 2024
Merged
Add MLFloat16 support for LayerNormalization, SkipLayerNormalization#22063amarin16 merged 20 commits intomicrosoft:mainfrom
amarin16 merged 20 commits intomicrosoft:mainfrom
Conversation
Contributor
|
Could it be faster to use MlasConvertHalfToFloatBuffer to convert inputs? |
Contributor
Author
@tianleiwu From what I can see, the logic inside |
Contributor
That's slow path. The fast path uses assembly kernel of AVX_NE_CONVERT instructions, which might be faster (It is not guarantee since there is extra I/O if we use temp buffers to hold the casted inputs. May need run benchmark to see whether it could help). |
Contributor
|
Please fix format like the following: |
Contributor
|
/azp run Linux CPU CI Pipeline, Windows CPU CI Pipeline |
|
Azure Pipelines successfully started running 2 pipeline(s). |
Contributor
|
SkipLayerNormTest.SkipLayerNormBatch1 unit test failed: |
xadupre
reviewed
Sep 12, 2024
xadupre
reviewed
Sep 12, 2024
xadupre
reviewed
Sep 12, 2024
xadupre
reviewed
Sep 12, 2024
xadupre
reviewed
Sep 12, 2024
Contributor
Author
Seems to be passing in the latest build, as well as locally |
Member
|
It seems to work. You still need to update the markdown pages for the documentation. You can generate them again or manually fix the differences by looking at the job output which makes a diff between the current version and the automatically generated one: https://dev.azure.com/onnxruntime/onnxruntime/_build/results?buildId=1491838&view=logs&j=7f366e99-16b2-52cc-e1ff-653af284e397&t=5c9cf234-f957-5fdc-9c37-89899bf73c0c&l=58. |
yufenglee
reviewed
Sep 18, 2024
yufenglee
approved these changes
Sep 23, 2024
Member
|
How is the perf comparing with fp32 version? |
amarin16
added a commit
that referenced
this pull request
Oct 15, 2024
- Added a microbenchmark for the `LayerNormalization` MLFloat16 support added in #22063. - Updated the `LayerNormalization` MLFloat16 implementation to improve the latency. ``` ---------------------------------------------------------------------------------------------- Original MLFloat16 support Time CPU Iterations ---------------------------------------------------------------------------------------------- BM_LayerNormalization<MLFloat16, float>/1/real_time 15599 us 15625 us 47 BM_LayerNormalization<MLFloat16, float>/1/real_time 14714 us 14824 us 39 BM_LayerNormalization<MLFloat16, float>/1/real_time 14634 us 14688 us 50 ---------------------------------------------------------------------------------------------- Updated MLFloat16 support Time CPU Iterations ---------------------------------------------------------------------------------------------- BM_LayerNormalization<MLFloat16, float>/1/real_time 7276 us 7254 us 84 BM_LayerNormalization<MLFloat16, float>/1/real_time 6820 us 6720 us 93 BM_LayerNormalization<MLFloat16, float>/1/real_time 6840 us 6882 us 84 ```
guschmue
pushed a commit
that referenced
this pull request
Oct 18, 2024
- Added a microbenchmark for the `LayerNormalization` MLFloat16 support added in #22063. - Updated the `LayerNormalization` MLFloat16 implementation to improve the latency. ``` ---------------------------------------------------------------------------------------------- Original MLFloat16 support Time CPU Iterations ---------------------------------------------------------------------------------------------- BM_LayerNormalization<MLFloat16, float>/1/real_time 15599 us 15625 us 47 BM_LayerNormalization<MLFloat16, float>/1/real_time 14714 us 14824 us 39 BM_LayerNormalization<MLFloat16, float>/1/real_time 14634 us 14688 us 50 ---------------------------------------------------------------------------------------------- Updated MLFloat16 support Time CPU Iterations ---------------------------------------------------------------------------------------------- BM_LayerNormalization<MLFloat16, float>/1/real_time 7276 us 7254 us 84 BM_LayerNormalization<MLFloat16, float>/1/real_time 6820 us 6720 us 93 BM_LayerNormalization<MLFloat16, float>/1/real_time 6840 us 6882 us 84 ```
rohan11235813
pushed a commit
to quadric-io/onnxruntime
that referenced
this pull request
Aug 19, 2025
- Added a microbenchmark for the `LayerNormalization` MLFloat16 support added in microsoft/onnxruntime#22063. - Updated the `LayerNormalization` MLFloat16 implementation to improve the latency. ``` ---------------------------------------------------------------------------------------------- Original MLFloat16 support Time CPU Iterations ---------------------------------------------------------------------------------------------- BM_LayerNormalization<MLFloat16, float>/1/real_time 15599 us 15625 us 47 BM_LayerNormalization<MLFloat16, float>/1/real_time 14714 us 14824 us 39 BM_LayerNormalization<MLFloat16, float>/1/real_time 14634 us 14688 us 50 ---------------------------------------------------------------------------------------------- Updated MLFloat16 support Time CPU Iterations ---------------------------------------------------------------------------------------------- BM_LayerNormalization<MLFloat16, float>/1/real_time 7276 us 7254 us 84 BM_LayerNormalization<MLFloat16, float>/1/real_time 6820 us 6720 us 93 BM_LayerNormalization<MLFloat16, float>/1/real_time 6840 us 6882 us 84 ```
rohan11235813
pushed a commit
to quadric-io/onnxruntime
that referenced
this pull request
Sep 15, 2025
- Added a microbenchmark for the `LayerNormalization` MLFloat16 support added in microsoft/onnxruntime#22063. - Updated the `LayerNormalization` MLFloat16 implementation to improve the latency. ``` ---------------------------------------------------------------------------------------------- Original MLFloat16 support Time CPU Iterations ---------------------------------------------------------------------------------------------- BM_LayerNormalization<MLFloat16, float>/1/real_time 15599 us 15625 us 47 BM_LayerNormalization<MLFloat16, float>/1/real_time 14714 us 14824 us 39 BM_LayerNormalization<MLFloat16, float>/1/real_time 14634 us 14688 us 50 ---------------------------------------------------------------------------------------------- Updated MLFloat16 support Time CPU Iterations ---------------------------------------------------------------------------------------------- BM_LayerNormalization<MLFloat16, float>/1/real_time 7276 us 7254 us 84 BM_LayerNormalization<MLFloat16, float>/1/real_time 6820 us 6720 us 93 BM_LayerNormalization<MLFloat16, float>/1/real_time 6840 us 6882 us 84 ```
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 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.
Add
MLFloat16support for:LayerNormalizationSimplifiedLayerNormalizationSkipLayerNormalizationSkipSimplifiedLayerNormalizationThere are existing
LayerNormTestunit tests that cover theMLFloat16functionality forLayerNormalizationonceMLFloat16is registered (for exampleLayerNormTest.LayerNorm_Scale_Float16Input).Similarly, there are unit tests such as
SkipLayerNormTest.SkipLayerNormBatch1_Float16that cover MLFloat16 inputs forSkipLayerNormalization.