[quant][pt2] Fix QAT convert for mobilenetv2 #104110
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/104110
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 New FailureAs of commit e759384: This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
This pull request was exported from Phabricator. Differential Revision: D46750343 |
andrewor14
force-pushed
the
export-D46750343
branch
from
96a5d29 to
c0d8a58
Compare
|
This pull request was exported from Phabricator. Differential Revision: D46750343 |
|
This pull request was exported from Phabricator. Differential Revision: D46750343 |
andrewor14
force-pushed
the
export-D46750343
branch
from
c0d8a58 to
386e0d6
Compare
andrewor14
force-pushed
the
export-D46750343
branch
from
386e0d6 to
1320255
Compare
|
This pull request was exported from Phabricator. Differential Revision: D46750343 |
andrewor14
force-pushed
the
export-D46750343
branch
from
1320255 to
53e5d32
Compare
|
This pull request was exported from Phabricator. Differential Revision: D46750343 |
andrewor14
force-pushed
the
export-D46750343
branch
from
53e5d32 to
4712d6f
Compare
|
This pull request was exported from Phabricator. Differential Revision: D46750343 |
andrewor14
force-pushed
the
export-D46750343
branch
from
4712d6f to
5fc046d
Compare
|
This pull request was exported from Phabricator. Differential Revision: D46750343 |
andrewor14
force-pushed
the
export-D46750343
branch
from
5fc046d to
1229942
Compare
|
This pull request was exported from Phabricator. Differential Revision: D46750343 |
Summary: Pull Request resolved: pytorch#104110 QAT convert for mobilenetv2 was previously not working because we incorrectly applied dropout during eval as well as training. This is because, for exported models, model.eval() does not change the behavior of dropout, unlike models with torch ops. This commit simulates the effects of model.eval() for exported models as well by replacing the aten dropout pattern before eval. As of this commit, end-to-end QAT numerics now match for mobilenetv2 between FX and PT2. Test Plan: python test/test_quantization.py TestQuantizePT2EModels.test_qat_mobilenet_v2 Reviewed By: jerryzh168 Differential Revision: D46750343 fbshipit-source-id: 38a1a027ed8680edc022a4c5a4015c3b5c811438
andrewor14
force-pushed
the
export-D46750343
branch
from
1229942 to
c489280
Compare
|
This pull request was exported from Phabricator. Differential Revision: D46750343 |
andrewor14
force-pushed
the
export-D46750343
branch
from
c489280 to
a42884f
Compare
|
This pull request was exported from Phabricator. Differential Revision: D46750343 |
1 similar comment
|
This pull request was exported from Phabricator. Differential Revision: D46750343 |
andrewor14
force-pushed
the
export-D46750343
branch
from
a42884f to
a9879f7
Compare
andrewor14
force-pushed
the
export-D46750343
branch
from
a9879f7 to
7ad12b1
Compare
|
This pull request was exported from Phabricator. Differential Revision: D46750343 |
Summary: Pull Request resolved: pytorch#104110 QAT convert for mobilenetv2 was previously not working because we incorrectly applied dropout during eval as well as training. This is because, for exported models, model.eval() does not change the behavior of dropout, unlike models with torch ops. This commit simulates the effects of model.eval() for exported models as well by replacing the aten dropout pattern before eval. As of this commit, end-to-end QAT numerics now match for mobilenetv2 between FX and PT2. Test Plan: python test/test_quantization.py TestQuantizePT2EModels.test_qat_mobilenet_v2 Reviewed By: jerryzh168 Differential Revision: D46750343 fbshipit-source-id: dcf508c0167c1c362410ca860824299c4b35bab6
andrewor14
force-pushed
the
export-D46750343
branch
from
7ad12b1 to
e759384
Compare
|
This pull request was exported from Phabricator. Differential Revision: D46750343 |
|
@pytorchbot merge |
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Summary:
QAT convert for mobilenetv2 was previously not working
because we incorrectly applied dropout during eval as well as
training. This is because, for exported models, model.eval() does
not change the behavior of dropout, unlike models with torch ops.
This commit simulates the effects of model.eval() for exported
models as well by replacing the aten dropout pattern before eval.
As of this commit, end-to-end QAT numerics now match for
mobilenetv2 between FX and PT2.
Test Plan: python test/test_quantization.py TestQuantizePT2EModels.test_qat_mobilenet_v2
Differential Revision: D46750343
cc @voznesenskym @penguinwu @anijain2305 @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @Xia-Weiwen @wenzhe-nrv @jiayisunx @ipiszy @chenyang78