ROCmSoftwarePlatform
Remove layernorm fusion
Codecov Report
All modified and coverable lines are covered by tests :white_check_mark:
Project coverage is 92.19%. Comparing base (
da545d2) to head (b9edc8e).
Additional details and impacted files
@@ Coverage Diff @@
## develop #3465 +/- ##
========================================
Coverage 92.19% 92.19%
========================================
Files 513 513
Lines 21633 21633
========================================
Hits 19945 19945
Misses 1688 1688
: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}
Even with #3659, the flux model doesnt give a proper output when using MIGRAPHX_DISABLE_LAYERNORM_FUSION=1. Need to resolve that before we can remove this
| Test | Batch | Rate new b9edc8 |
Rate old da545d |
Diff | Compare |
|---|---|---|---|---|---|
| torchvision-resnet50 | 64 | 3,256.83 | 3,255.84 | 0.03% | :white_check_mark: |
| torchvision-resnet50_fp16 | 64 | 6,984.97 | 6,991.64 | -0.10% | :white_check_mark: |
| torchvision-densenet121 | 32 | 2,434.77 | 2,431.85 | 0.12% | :white_check_mark: |
| torchvision-densenet121_fp16 | 32 | 4,086.28 | 4,070.27 | 0.39% | :white_check_mark: |
| torchvision-inceptionv3 | 32 | 1,629.60 | 1,628.09 | 0.09% | :white_check_mark: |
| torchvision-inceptionv3_fp16 | 32 | 2,744.33 | 2,747.87 | -0.13% | :white_check_mark: |
| cadene-inceptionv4 | 16 | 765.69 | 764.69 | 0.13% | :white_check_mark: |
| cadene-resnext64x4 | 16 | 810.56 | 806.89 | 0.45% | :white_check_mark: |
| slim-mobilenet | 64 | 7,466.19 | 7,464.92 | 0.02% | :white_check_mark: |
| slim-nasnetalarge | 64 | 208.46 | 208.40 | 0.03% | :white_check_mark: |
| slim-resnet50v2 | 64 | 3,440.71 | 3,441.42 | -0.02% | :white_check_mark: |
| bert-mrpc-onnx | 8 | 1,149.67 | 1,145.08 | 0.40% | :white_check_mark: |
| bert-mrpc-tf | 1 | 464.16 | 461.72 | 0.53% | :white_check_mark: |
| pytorch-examples-wlang-gru | 1 | 418.25 | 429.74 | -2.67% | :white_check_mark: |
| pytorch-examples-wlang-lstm | 1 | 389.51 | 480.37 | -18.91% | :red_circle: |
| torchvision-resnet50_1 | 1 | 786.96 | 770.17 | 2.18% | :white_check_mark: |
| cadene-dpn92_1 | 1 | 404.92 | 403.49 | 0.36% | :white_check_mark: |
| cadene-resnext101_1 | 1 | 382.37 | 381.97 | 0.10% | :white_check_mark: |
| onnx-taau-downsample | 1 | 345.60 | 346.00 | -0.12% | :white_check_mark: |
| dlrm-criteoterabyte | 1 | 33.32 | 33.31 | 0.02% | :white_check_mark: |
| dlrm-criteoterabyte_fp16 | 1 | 52.77 | 52.72 | 0.09% | :white_check_mark: |
| agentmodel | 1 | 7,985.43 | 8,212.54 | -2.77% | :white_check_mark: |
| unet_fp16 | 2 | 57.01 | 58.71 | -2.90% | :white_check_mark: |
| resnet50v1_fp16 | 1 | 1,022.58 | 938.46 | 8.96% | :high_brightness: |
| resnet50v1_int8 | 1 | 1,009.23 | 999.92 | 0.93% | :white_check_mark: |
| bert_base_cased_fp16 | 64 | 1,168.53 | 1,169.75 | -0.10% | :white_check_mark: |
| bert_large_uncased_fp16 | 32 | 363.40 | 363.16 | 0.06% | :white_check_mark: |
| bert_large_fp16 | 1 | 203.99 | 200.11 | 1.94% | :white_check_mark: |
| distilgpt2_fp16 | 16 | 2,211.12 | 2,198.98 | 0.55% | :white_check_mark: |
| yolov5s | 1 | 530.23 | 533.15 | -0.55% | :white_check_mark: |
| tinyllama | 1 | 43.46 | 43.41 | 0.10% | :white_check_mark: |
| vicuna-fastchat | 1 | 175.67 | 173.49 | 1.26% | :white_check_mark: |
| whisper-tiny-encoder | 1 | 417.70 | 417.39 | 0.07% | :white_check_mark: |
| whisper-tiny-decoder | 1 | 427.57 | 427.99 | -0.10% | :white_check_mark: |
This build is not recommended to merge :red_circle:
:red_circle:bert_large_uncased_fp16: FAILED: MIGraphX is not within tolerance - check verbose output