ROCmSoftwarePlatform
Set warmup and tuning iterations through env variable for hipBLASLt
This is a draft PR with changes to test different number of warmup and tuning iterations for hipBLASLt.
Codecov Report
All modified and coverable lines are covered by tests :white_check_mark:
Additional details and impacted files
@@ Coverage Diff @@
## develop #3792 +/- ##
===========================================
+ Coverage 92.30% 92.53% +0.23%
===========================================
Files 519 520 +1
Lines 22233 22747 +514
===========================================
+ Hits 20520 21047 +527
+ Misses 1713 1700 -13
:rocket: New features to boost your workflow:
- :snowflake: Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
![codecov[bot] avatar](https://avatars.githubusercontent.com/in/254?v=4 "Published by a pub.dev verified publisher"){kind=small}
| Test | Batch | Rate new 50ec9a |
Rate old 5dc019 |
Diff | Compare |
|---|---|---|---|---|---|
| torchvision-resnet50 | 64 | 3,233.07 | 3,229.26 | 0.12% | :white_check_mark: |
| torchvision-resnet50_fp16 | 64 | 6,864.21 | 6,862.73 | 0.02% | :white_check_mark: |
| torchvision-densenet121 | 32 | 2,430.89 | 2,429.91 | 0.04% | :white_check_mark: |
| torchvision-densenet121_fp16 | 32 | 4,167.69 | 4,173.61 | -0.14% | :white_check_mark: |
| torchvision-inceptionv3 | 32 | 1,612.25 | 1,611.02 | 0.08% | :white_check_mark: |
| torchvision-inceptionv3_fp16 | 32 | 2,683.80 | 2,683.26 | 0.02% | :white_check_mark: |
| cadene-inceptionv4 | 16 | 749.16 | 749.06 | 0.01% | :white_check_mark: |
| cadene-resnext64x4 | 16 | 808.50 | 808.64 | -0.02% | :white_check_mark: |
| slim-mobilenet | 64 | 6,659.37 | 6,657.58 | 0.03% | :white_check_mark: |
| slim-nasnetalarge | 64 | 198.93 | 198.90 | 0.01% | :white_check_mark: |
| slim-resnet50v2 | 64 | 3,429.01 | 3,424.77 | 0.12% | :white_check_mark: |
| bert-mrpc-onnx | 8 | 1,139.23 | 1,138.15 | 0.09% | :white_check_mark: |
| bert-mrpc-tf | 1 | 470.41 | 470.57 | -0.04% | :white_check_mark: |
| pytorch-examples-wlang-gru | 1 | 426.61 | 425.82 | 0.18% | :white_check_mark: |
| pytorch-examples-wlang-lstm | 1 | 460.12 | 397.58 | 15.73% | :high_brightness: |
| torchvision-resnet50_1 | 1 | 784.25 | 789.46 | -0.66% | :white_check_mark: |
| cadene-dpn92_1 | 1 | 416.68 | 415.00 | 0.41% | :white_check_mark: |
| cadene-resnext101_1 | 1 | 389.67 | 389.35 | 0.08% | :white_check_mark: |
| onnx-taau-downsample | 1 | 373.16 | 372.43 | 0.20% | :white_check_mark: |
| dlrm-criteoterabyte | 1 | 30.55 | 30.53 | 0.05% | :white_check_mark: |
| dlrm-criteoterabyte_fp16 | 1 | 49.15 | 49.09 | 0.12% | :white_check_mark: |
| agentmodel | 1 | 7,745.75 | 7,273.63 | 6.49% | :high_brightness: |
| unet_fp16 | 2 | 57.74 | 57.82 | -0.15% | :white_check_mark: |
| resnet50v1_fp16 | 1 | 957.53 | 978.89 | -2.18% | :white_check_mark: |
| resnet50v1_int8 | 1 | 787.43 | 781.27 | 0.79% | :white_check_mark: |
| bert_base_cased_fp16 | 64 | 1,172.27 | 1,171.32 | 0.08% | :white_check_mark: |
| bert_large_uncased_fp16 | 32 | 362.45 | 362.15 | 0.08% | :white_check_mark: |
| bert_large_fp16 | 1 | 199.68 | 198.55 | 0.57% | :white_check_mark: |
| distilgpt2_fp16 | 16 | 2,214.44 | 2,214.29 | 0.01% | :white_check_mark: |
| yolov5s | 1 | 527.42 | 515.26 | 2.36% | :white_check_mark: |
| tinyllama | 1 | 43.46 | 43.42 | 0.08% | :white_check_mark: |
| vicuna-fastchat | 1 | 43.75 | 43.81 | -0.14% | :white_check_mark: |
| whisper-tiny-encoder | 1 | 410.78 | 410.65 | 0.03% | :white_check_mark: |
| whisper-tiny-decoder | 1 | 407.57 | 406.71 | 0.21% | :white_check_mark: |
Check results before merge :high_brightness:
:red_circle:bert_large_uncased_fp16: FAILED: MIGraphX is not within tolerance - check verbose output
#3955 Implements changes that can be used to change the number of tuning iterations. Cold iterations are set to 1, and cannot be changed with the current implementation since there is no evidence to suggest that changing the number of cold iterations makes a difference. Closing this PR since the changes in this PR are no longer relevant.