[quant][graphmode] Skip quantizing input and output in matched module by jerryzh168 · Pull Request #32814 · pytorch/pytorch

jerryzh168 · 2020-01-30T02:59:36Z

Stack from ghstack:

[not4land] debugging #33650 [not4land] debugging
[not4land] Temporary workaround to get quantized ops #33533 [not4land] Temporary workaround to get quantized ops
[quant][graphmode] insert prepack/unpack for aten::linear #33895 [quant][graphmode] insert prepack/unpack for aten::linear
[quant][graphmode] Quantization pattern for aten::linear #33854 [quant][graphmode] Quantization pattern for aten::linear
[quant][graphmode] Add quantization pattern for quantized::add_relu #33532 [quant][graphmode] Add quantization pattern for quantized::add_relu
[quant][graphmode] Add quantized conv2d-relu fusion pattern #33279 [quant][graphmode] Add quantized conv2d-relu fusion pattern
[quant][graphmode] Add Finalize function that inlines graph and produce quantized ops #33927 [quant][graphmode] Add Finalize function that inlines graph and produce quantized ops
[quant][graphmode] Swap quantized functional linear with aten::linear #33853 [quant][graphmode] Swap quantized functional linear with aten::linear
[quant][graphmode] Handling ops doesn't require observation in insertObservers #33481 [quant][graphmode] Handling ops doesn't require observation in insertObservers
[quant][graphmode] Add add_relu pattern in skip values #32816 [quant][graphmode] Add add_relu pattern in skip values
[quant][graphmode] Skip quantizing input and output in matched module #32814 [quant][graphmode] Skip quantizing input and output in matched module

Summary:
We skip quantization for the intermediate values for patterns like Conv - ReLU,
but currently we didn't skip quantizing the input/output of the graphs of matched modules,
since we now changed the way we add observers, this also needs to be updated.

Test Plan:
python test/test_jit.py -- 'TestJit.test_insert_observers_skip_values'
Reviewers:
mvz

Subscribers:

Tasks:

Tags:

Differential Revision: D20208785

jerryzh168 · 2020-01-30T03:03:49Z

description needs to be updated

kostmo · 2020-01-30T03:04:00Z

💊 CircleCI build failures summary and remediations

As of commit 5251af4 (more details on the Dr. CI page):

✅ None of the build failures appear to be your fault 💚

4/6 tentatively recognized as flaky ❄️
- Click here to rerun these jobs
2/6 broken upstream at merge base f299c2d since Mar 03
Please rebase on the viable/strict branch (expand for instructions)

If your commit is newer than viable/strict, you can try basing on an older, stable commit:
```
git fetch origin viable/strict
git rebase --onto viable/strict $(git merge-base origin/master HEAD)
```
If your commit is older than viable/strict:
```
git fetch origin viable/strict
git rebase viable/strict
```
Check out the recency history of this "viable master" tracking branch.

❄️ 4 tentatively flaky failures

4 failures tentatively classified as flaky but have not launched reruns to confirm:

pytorch_linux_xenial_py3_6_gcc5_4_test (1/4)

Step: "Test" (full log | pattern match details) ❄️

Mar 03 23:45:47 RuntimeError: Error downloading resource!

Mar 03 23:45:47  
Mar 03 23:45:47 During handling of the above exception, another exception occurred: 
Mar 03 23:45:47  
Mar 03 23:45:47 Traceback (most recent call last): 
Mar 03 23:45:47   File "tools/download_mnist.py", line 87, in <module> 
Mar 03 23:45:47     main() 
Mar 03 23:45:47   File "tools/download_mnist.py", line 80, in main 
Mar 03 23:45:47     download(path, url, options.quiet) 
Mar 03 23:45:47   File "tools/download_mnist.py", line 41, in download 
Mar 03 23:45:47     raise RuntimeError('Error downloading resource!') 
Mar 03 23:45:47 RuntimeError: Error downloading resource! 
Mar 03 23:45:47 + cleanup 
Mar 03 23:45:47 + retcode=1 
Mar 03 23:45:47 + set +x 
Mar 03 23:45:47 =================== sccache compilation log =================== 
Mar 03 23:45:47 ERROR:sccache::server: Compilation failed: Output { status: ExitStatus(ExitStatus(256)), stdout: "", stderr: "/tmp/torch_extensions/test_compilation_error_formatting/main.cpp: In function \'int main()\':\n/tmp/torch_extensions/test_compilation_error_formatting/main.cpp:2:23: error: expected \';\' before \'}\' token\n int main() { return 0 }\n                       ^\n" } 
Mar 03 23:45:47  
Mar 03 23:45:47 =========== If your build fails, please take a look at the log above for possible reasons =========== 
Mar 03 23:45:47 Compile requests                 68 
Mar 03 23:45:47 Compile requests executed        28 
Mar 03 23:45:47 Cache hits                       21

pytorch_linux_xenial_cuda10_1_cudnn7_py3_NO_AVX_NO_AVX2_test (2/4)

Step: "Test" (full log | pattern match details) ❄️

Mar 04 00:33:36 RuntimeError: Error downloading resource!

Mar 04 00:33:36  
Mar 04 00:33:36 During handling of the above exception, another exception occurred: 
Mar 04 00:33:36  
Mar 04 00:33:36 Traceback (most recent call last): 
Mar 04 00:33:36   File "tools/download_mnist.py", line 87, in <module> 
Mar 04 00:33:36     main() 
Mar 04 00:33:36   File "tools/download_mnist.py", line 80, in main 
Mar 04 00:33:36     download(path, url, options.quiet) 
Mar 04 00:33:36   File "tools/download_mnist.py", line 41, in download 
Mar 04 00:33:36     raise RuntimeError('Error downloading resource!') 
Mar 04 00:33:36 RuntimeError: Error downloading resource! 
Mar 04 00:33:36 + cleanup 
Mar 04 00:33:36 + retcode=1 
Mar 04 00:33:36 + set +x 
Mar 04 00:33:36 =================== sccache compilation log =================== 
Mar 04 00:33:36 =========== If your build fails, please take a look at the log above for possible reasons =========== 
Mar 04 00:33:36 Compile requests                19 
Mar 04 00:33:36 Compile requests executed        0 
Mar 04 00:33:36 Cache hits                       0 
Mar 04 00:33:36 Cache misses                     0 
Mar 04 00:33:36 Cache timeouts                   0

pytorch_linux_xenial_cuda10_1_cudnn7_py3_gcc7_test (3/4)

Step: "Test" (full log | pattern match details) ❄️

Mar 04 00:36:42 RuntimeError: Error downloading resource!

Mar 04 00:36:42  
Mar 04 00:36:42 During handling of the above exception, another exception occurred: 
Mar 04 00:36:42  
Mar 04 00:36:42 Traceback (most recent call last): 
Mar 04 00:36:42   File "tools/download_mnist.py", line 87, in <module> 
Mar 04 00:36:42     main() 
Mar 04 00:36:42   File "tools/download_mnist.py", line 80, in main 
Mar 04 00:36:42     download(path, url, options.quiet) 
Mar 04 00:36:42   File "tools/download_mnist.py", line 41, in download 
Mar 04 00:36:42     raise RuntimeError('Error downloading resource!') 
Mar 04 00:36:42 RuntimeError: Error downloading resource! 
Mar 04 00:36:42 + cleanup 
Mar 04 00:36:42 + retcode=1 
Mar 04 00:36:42 + set +x 
Mar 04 00:36:42 =================== sccache compilation log =================== 
Mar 04 00:36:42 =========== If your build fails, please take a look at the log above for possible reasons =========== 
Mar 04 00:36:42 Compile requests                19 
Mar 04 00:36:42 Compile requests executed        0 
Mar 04 00:36:42 Cache hits                       0 
Mar 04 00:36:42 Cache misses                     0 
Mar 04 00:36:42 Cache timeouts                   0

pytorch_linux_xenial_py3_clang5_asan_test (4/4)

Step: "Test" (full log | pattern match details) ❄️

Mar 04 01:15:01 RuntimeError: Error downloading resource!

Mar 04 01:15:01  
Mar 04 01:15:01 During handling of the above exception, another exception occurred: 
Mar 04 01:15:01  
Mar 04 01:15:01 Traceback (most recent call last): 
Mar 04 01:15:01   File "tools/download_mnist.py", line 87, in <module> 
Mar 04 01:15:01     main() 
Mar 04 01:15:01   File "tools/download_mnist.py", line 80, in main 
Mar 04 01:15:01     download(path, url, options.quiet) 
Mar 04 01:15:01   File "tools/download_mnist.py", line 41, in download 
Mar 04 01:15:01     raise RuntimeError('Error downloading resource!') 
Mar 04 01:15:01 RuntimeError: Error downloading resource! 
Mar 04 01:15:01 + cleanup 
Mar 04 01:15:01 + retcode=1 
Mar 04 01:15:01 + set +x 
Mar 04 01:15:01 =================== sccache compilation log =================== 
Mar 04 01:15:01 =========== If your build fails, please take a look at the log above for possible reasons =========== 
Mar 04 01:15:01 Compile requests                13 
Mar 04 01:15:01 Compile requests executed        0 
Mar 04 01:15:01 Cache hits                       0 
Mar 04 01:15:01 Cache misses                     0 
Mar 04 01:15:01 Cache timeouts                   0

🚧 2 upstream failures recognized by patterns:

These were probably caused by upstream breakages:

pytorch_linux_xenial_cuda10_1_cudnn7_py3_nogpu_test from Mar 03
pytorch_xla_linux_xenial_py3_6_clang7_test from Mar 03

This comment was automatically generated by Dr. CI (expand for details).

Follow this link to opt-out of these comments for your Pull Requests.

Please report bugs/suggestions on the GitHub issue tracker.

This comment has been revised 176 times.

…ched module" Summary: We skip quantization for the intermediate values for patterns like `Conv - ReLU`, but currently we didn't skip quantizing the input/output of the graphs of matched modules, since we now changed the way we add observers, this also needs to be updated. Test Plan: python test/test_jit.py -- 'TestJit.test_insert_observers_skip_values' Reviewers: mvz Subscribers: Tasks: Tags: [ghstack-poisoned]

…ched module" Summary: We skip quantization for the intermediate values for patterns like `Conv - ReLU`, but currently we didn't skip quantizing the input/output of the graphs of matched modules, since we now changed the way we add observers, this also needs to be updated. Test Plan: python test/test_jit.py -- 'TestJit.test_insert_observers_skip_values' Reviewers: mvz Subscribers: Tasks: Tags: Differential Revision: [D20208785](https://our.internmc.facebook.com/intern/diff/D20208785) [ghstack-poisoned]

facebook-github-bot · 2020-03-04T05:16:28Z

This pull request has been merged in e5bbd23.

ljk53 · 2020-03-04T05:46:44Z

@jerryzh168 do you think this might be related to the master CI failure: https://app.circleci.com/jobs/github/pytorch/pytorch/4684269 ?

This PR stacked with my PR #34055 so it's not obvious which one is the cause, but mine shouldn't affect test_jit :) cc: @mrshenli

jerryzh168 · 2020-03-04T05:51:37Z

This PR is not changing jit and test_script_pack_padded_sequence is not a quantization test.
Is this failing consistently?

…d module (#32814)" This reverts commit e5bbd23. [ghstack-poisoned]

…d module (#32814)" This reverts commit e5bbd23. ghstack-source-id: 19a0174 Pull Request resolved: #34188

…pytorch#32814) Summary: Pull Request resolved: pytorch#32814 We skip quantization for the intermediate values for patterns like `Conv - ReLU`, but currently we didn't skip quantizing the input/output of the graphs of matched modules, since we now changed the way we add observers, this also needs to be updated. Test Plan: python test/test_jit.py -- 'TestJit.test_insert_observers_skip_values' Imported from OSS Differential Revision: D20208785 fbshipit-source-id: ce30f2c4c8ce737500d0b41357c80ec8b33aecf9

jerryzh168 requested a review from apaszke as a code owner January 30, 2020 02:59

facebook-github-bot added the oncall: jit Add this issue/PR to JIT oncall triage queue label Jan 30, 2020

jerryzh168 requested a review from ZolotukhinM January 30, 2020 03:04

This was referenced Jan 30, 2020

[jit][fix] Remove slot in parameter slot #32846

Closed

Add type annotation for bias in _ConvNd #32885

Closed

jerryzh168 changed the title ~~[quant][graphmode] Skip quantizing input in matched module~~ [quant][graphmode] Skip quantizing input and output in matched module Jan 31, 2020

jerryzh168 mentioned this pull request Feb 4, 2020

[jit] Add None parameter as parameter instead of attributes #32964

Closed

This was referenced Feb 27, 2020

[quant][graphmode] Swap quantized functional linear with aten::linear #33853

Closed

[quant][graphmode] Quantization pattern for aten::linear #33854

Closed

jerryzh168 mentioned this pull request Feb 27, 2020

[quant][graphmode] insert prepack/unpack for aten::linear #33895

Closed

raghuramank100 approved these changes Feb 28, 2020

View reviewed changes

jerryzh168 mentioned this pull request Feb 28, 2020

[quant][graphmode] Add Finalize function that inlines graph and produce quantized ops #33927

Closed

jerryzh168 added 7 commits February 27, 2020 22:46

facebook-github-bot closed this in e5bbd23 Mar 4, 2020

facebook-github-bot added the merged label Mar 4, 2020

ljk53 added a commit that referenced this pull request Mar 4, 2020

Revert "[quant][graphmode] Skip quantizing input and output in matche…

278da51

…d module (#32814)" This reverts commit e5bbd23. [ghstack-poisoned]

ljk53 mentioned this pull request Mar 4, 2020

Revert "[quant][graphmode] Skip quantizing input and output in matched module (#32814)" #34188

Closed

ljk53 added a commit that referenced this pull request Mar 4, 2020

Revert "[quant][graphmode] Skip quantizing input and output in matche…

243a3cd

…d module (#32814)" This reverts commit e5bbd23. ghstack-source-id: 19a0174 Pull Request resolved: #34188

ljk53 mentioned this pull request Mar 4, 2020

[pytorch][mobile] support for custom mobile build with dynamic dispatch #34055

Closed

mrshenli mentioned this pull request Mar 4, 2020

DISABLED test_script_pack_padded_sequence (test_jit.TestScript) #34213

Closed

facebook-github-bot deleted the gh/jerryzh168/204/head branch March 7, 2020 15:18

mruberry added the Merged label Oct 28, 2020

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[quant][graphmode] Skip quantizing input and output in matched module#32814

[quant][graphmode] Skip quantizing input and output in matched module#32814
jerryzh168 wants to merge 19 commits intogh/jerryzh168/204/basefrom
gh/jerryzh168/204/head

jerryzh168 commented Jan 30, 2020 •

edited

Loading

Uh oh!

jerryzh168 commented Jan 30, 2020

Uh oh!

kostmo commented Jan 30, 2020 •

edited by dr-ci Bot

Loading

Uh oh!

facebook-github-bot commented Mar 4, 2020

Uh oh!

ljk53 commented Mar 4, 2020

Uh oh!

jerryzh168 commented Mar 4, 2020

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

7 participants

Conversation

jerryzh168 commented Jan 30, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jerryzh168 commented Jan 30, 2020

Uh oh!

kostmo commented Jan 30, 2020 • edited by dr-ci Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

💊 CircleCI build failures summary and remediations

❄️ 4 tentatively flaky failures

pytorch_linux_xenial_py3_6_gcc5_4_test (1/4)

pytorch_linux_xenial_cuda10_1_cudnn7_py3_NO_AVX_NO_AVX2_test (2/4)

pytorch_linux_xenial_cuda10_1_cudnn7_py3_gcc7_test (3/4)

pytorch_linux_xenial_py3_clang5_asan_test (4/4)

🚧 2 upstream failures recognized by patterns:

Uh oh!

facebook-github-bot commented Mar 4, 2020

Uh oh!

ljk53 commented Mar 4, 2020

Uh oh!

jerryzh168 commented Mar 4, 2020

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

7 participants

jerryzh168 commented Jan 30, 2020 •

edited

Loading

kostmo commented Jan 30, 2020 •

edited by dr-ci Bot

Loading