ROCmSoftwarePlatform
Add ONNX parsing for SimplifiedLayerNormalization
Codecov Report
Attention: Patch coverage is 92.85714% with 2 lines in your changes missing coverage. Please review.
Project coverage is 92.26%. Comparing base (
e0a2325) to head (6f98953). Report is 151 commits behind head on develop.
| Files with missing lines | Patch % | Lines |
|---|---|---|
| src/onnx/parse_simplified_layer_normalization.cpp | 92.85% | 2 Missing :warning: |
Additional details and impacted files
@@ Coverage Diff @@
## develop #3129 +/- ##
========================================
Coverage 92.26% 92.26%
========================================
Files 499 500 +1
Lines 20020 20048 +28
========================================
+ Hits 18471 18497 +26
- Misses 1549 1551 +2
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
![codecov[bot] avatar](https://avatars.githubusercontent.com/in/254?v=4 "Published by a pub.dev verified publisher"){kind=small}
| Test | Batch | Rate new 8f10e7 |
Rate old b9cce0 |
Diff | Compare |
|---|---|---|---|---|---|
| torchvision-resnet50 | 64 | 1,713.97 | 1,713.09 | 0.05% | :white_check_mark: |
| torchvision-resnet50_fp16 | 64 | 3,810.29 | 3,810.54 | -0.01% | :white_check_mark: |
| torchvision-densenet121 | 32 | 1,455.43 | 1,453.33 | 0.14% | :white_check_mark: |
| torchvision-densenet121_fp16 | 32 | 2,431.94 | 2,431.93 | 0.00% | :white_check_mark: |
| torchvision-inceptionv3 | 32 | 882.48 | 883.38 | -0.10% | :white_check_mark: |
| torchvision-inceptionv3_fp16 | 32 | 1,416.07 | 1,414.61 | 0.10% | :white_check_mark: |
| cadene-inceptionv4 | 16 | 407.52 | 407.58 | -0.01% | :white_check_mark: |
| cadene-resnext64x4 | 16 | 413.54 | 413.72 | -0.05% | :white_check_mark: |
| slim-mobilenet | 64 | 3,822.77 | 3,822.77 | 0.00% | :white_check_mark: |
| slim-nasnetalarge | 64 | 97.03 | 97.02 | 0.00% | :white_check_mark: |
| slim-resnet50v2 | 64 | 1,650.65 | 1,651.90 | -0.08% | :white_check_mark: |
| bert-mrpc-onnx | 8 | 589.47 | 591.41 | -0.33% | :white_check_mark: |
| bert-mrpc-tf | 1 | 289.00 | 289.91 | -0.31% | :white_check_mark: |
| pytorch-examples-wlang-gru | 1 | 335.15 | 332.28 | 0.86% | :white_check_mark: |
| pytorch-examples-wlang-lstm | 1 | 298.82 | 298.17 | 0.22% | :white_check_mark: |
| torchvision-resnet50_1 | 1 | 452.70 | 453.96 | -0.28% | :white_check_mark: |
| cadene-dpn92_1 | 1 | 244.70 | 244.78 | -0.03% | :white_check_mark: |
| cadene-resnext101_1 | 1 | 189.07 | 189.18 | -0.05% | :white_check_mark: |
| onnx-taau-downsample | 1 | 204.13 | 203.94 | 0.09% | :white_check_mark: |
| dlrm-criteoterabyte | 1 | 22.28 | 22.27 | 0.03% | :white_check_mark: |
| dlrm-criteoterabyte_fp16 | 1 | 41.65 | 41.62 | 0.07% | :white_check_mark: |
| agentmodel | 1 | 6,115.91 | 5,896.57 | 3.72% | :high_brightness: |
| unet_fp16 | 2 | 33.73 | 33.73 | 0.01% | :white_check_mark: |
| resnet50v1_fp16 | 1 | 561.82 | 564.20 | -0.42% | :white_check_mark: |
| resnet50v1_int8 | 1 | 463.79 | 462.59 | 0.26% | :white_check_mark: |
| bert_base_cased_fp16 | 64 | 620.74 | 620.70 | 0.01% | :white_check_mark: |
| bert_large_uncased_fp16 | 32 | 193.75 | 193.75 | 0.00% | :white_check_mark: |
| bert_large_fp16 | 1 | 103.89 | 103.85 | 0.05% | :white_check_mark: |
| distilgpt2_fp16 | 16 | 1,189.19 | 1,187.89 | 0.11% | :white_check_mark: |
| yolov5s | 1 | 297.21 | 298.42 | -0.41% | :white_check_mark: |
| tinyllama | 1 | 23.32 | 23.34 | -0.07% | :white_check_mark: |
| vicuna-fastchat | 1 | 132.70 | 134.00 | -0.98% | :white_check_mark: |
| whisper-tiny-encoder | 1 | 241.38 | 241.32 | 0.02% | :white_check_mark: |
| whisper-tiny-decoder | 1 | 245.53 | 245.93 | -0.16% | :white_check_mark: |
Check results before merge :high_brightness:
:red_circle:bert_large_uncased_fp16: FAILED: MIGraphX is not within tolerance - check verbose output
Where is this spec for this operator?
There isn't actually one for SimplifiedLayerNormalization, but this is the spec for SkipSimplifiedLayerNormalization, which is just add + SLN. That spec does include an optional bias input, but neither of the ORT implementations utilize it, so I omitted it from ours.
Looks fine, minor comments. I would like to see the equation that it's supposed to be doing.
The equation is the same as RMS LayerNorm.