Feature/vonmises upstream #33418

ahmadsalim · 2020-02-17T14:00:36Z

Third try of #33177 😄

dr-ci · 2020-02-17T14:07:02Z

💊 CircleCI build failures summary and remediations

As of commit a3dfc73:

1/1 failures introduced in this PR

Detailed failure analysis

One may explore the probable reasons each build failed interactively on the Dr. CI website.

🕵️ 1 new failure recognized by patterns

The following build failures do not appear to be due to upstream breakage:

pytorch_linux_xenial_py3_6_gcc5_4_test (1/1)

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

Feb 25 15:14:21 RuntimeError: test_quantization failed!

Feb 25 15:14:21 Ran 36 tests in 58.018s 
Feb 25 15:14:21  
Feb 25 15:14:21 FAILED (errors=1, skipped=1) 
Feb 25 15:14:21  
Feb 25 15:14:21 Generating XML reports... 
Feb 25 15:14:21 Traceback (most recent call last): 
Feb 25 15:14:21   File "test/run_test.py", line 493, in <module> 
Feb 25 15:14:21     main() 
Feb 25 15:14:21   File "test/run_test.py", line 486, in main 
Feb 25 15:14:21     raise RuntimeError(message) 
Feb 25 15:14:21 RuntimeError: test_quantization failed! 
Feb 25 15:14:21 + cleanup 
Feb 25 15:14:21 + retcode=1 
Feb 25 15:14:21 + set +x 
Feb 25 15:14:21 =================== sccache compilation log =================== 
Feb 25 15:14:21 =========== If your build fails, please take a look at the log above for possible reasons =========== 
Feb 25 15:14:21 Compile requests                  7 
Feb 25 15:14:21 Compile requests executed         6 
Feb 25 15:14:21 Cache hits                        0 
Feb 25 15:14:21 Cache misses                      6 
Feb 25 15:14:21 Cache timeouts                    0

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 8 times.

Add tests for VonMises logprob and sample Signed-off-by: Ahmad Salim Al-Sibahi <ahmad@di.ku.dk>

Fix linting issues and JIT compilation Fix proposal_r Fix issues with JIT not working with torch.Size

fritzo

LGTM after moving tests down to avoid rng seed churn

test/test_distributions.py

torch/distributions/von_mises.py

fritzo

LGTM

Test errors appear unrelated, all relating to TestFuser

======================================================================
ERROR: test_abs_cpu (__main__.TestFuser)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\Users\circleci\project\build\win_tmp\build\torch\testing\_internal\jit_utils.py", line 592, in wrapper
    fn(*args, **kwargs)
  File "test_jit_fuser.py", line 79, in test_abs_cpu
    self._test_fused_abs()
  File "test_jit_fuser.py", line 74, in _test_fused_abs
    self.assertAllFused(func.graph_for(a))
  File "C:\Users\circleci\project\build\win_tmp\build\torch\jit\__init__.py", line 2094, in _graph_for
    self(*args, **kwargs)
  File "C:\Users\circleci\project\build\win_tmp\build\torch\testing\_internal\common_utils.py", line 87, in prof_func_call
    return prof_callable(func_call, *args, **kwargs)
  File "C:\Users\circleci\project\build\win_tmp\build\torch\testing\_internal\common_utils.py", line 84, in prof_callable
    return callable(*args, **kwargs)
RuntimeError: Caught an unknown exception!

======================================================================
ERROR: test_chunk_correctness (__main__.TestFuser)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\Users\circleci\project\build\win_tmp\build\torch\testing\_internal\jit_utils.py", line 592, in wrapper
    fn(*args, **kwargs)
  File "test_jit_fuser.py", line 223, in test_chunk_correctness
    return self._test_chunk_correctness(self, 'cpu')
  File "test_jit_fuser.py", line 218, in _test_chunk_correctness
    self.checkScript(fn, [tensor])
  File "C:\Users\circleci\project\build\win_tmp\build\torch\testing\_internal\jit_utils.py", line 398, in checkScript
    opt_script_outputs = scripted_fn(*recording_inputs)
  File "C:\Users\circleci\project\build\win_tmp\build\torch\testing\_internal\common_utils.py", line 87, in prof_func_call
    return prof_callable(func_call, *args, **kwargs)
  File "C:\Users\circleci\project\build\win_tmp\build\torch\testing\_internal\common_utils.py", line 84, in prof_callable
    return callable(*args, **kwargs)
RuntimeError: Caught an unknown exception!

======================================================================
ERROR: test_scalar (__main__.TestFuser)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\Users\circleci\project\build\win_tmp\build\torch\testing\_internal\jit_utils.py", line 592, in wrapper
    fn(*args, **kwargs)
  File "test_jit_fuser.py", line 851, in test_scalar
    ge = self.checkScript(fn, (x, y))
  File "C:\Users\circleci\project\build\win_tmp\build\torch\testing\_internal\jit_utils.py", line 398, in checkScript
    opt_script_outputs = scripted_fn(*recording_inputs)
  File "C:\Users\circleci\project\build\win_tmp\build\torch\testing\_internal\common_utils.py", line 87, in prof_func_call
    return prof_callable(func_call, *args, **kwargs)
  File "C:\Users\circleci\project\build\win_tmp\build\torch\testing\_internal\common_utils.py", line 84, in prof_callable
    return callable(*args, **kwargs)
RuntimeError: Caught an unknown exception!

======================================================================
ERROR: test_where_and_typing (__main__.TestFuser)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\Users\circleci\project\build\win_tmp\build\torch\testing\_internal\jit_utils.py", line 592, in wrapper
    fn(*args, **kwargs)
  File "test_jit_fuser.py", line 906, in test_where_and_typing
    self.assertAllFused(script_f.graph_for(x, y), except_for={'prim::TupleConstruct'})
  File "C:\Users\circleci\project\build\win_tmp\build\torch\jit\__init__.py", line 2094, in _graph_for
    self(*args, **kwargs)
  File "C:\Users\circleci\project\build\win_tmp\build\torch\testing\_internal\common_utils.py", line 87, in prof_func_call
    return prof_callable(func_call, *args, **kwargs)
  File "C:\Users\circleci\project\build\win_tmp\build\torch\testing\_internal\common_utils.py", line 84, in prof_callable
    return callable(*args, **kwargs)
RuntimeError: Caught an unknown exception!

----------------------------------------------------------------------
Ran 46 tests in 12.462s

FAILED (errors=4, skipped=10)
Traceback (most recent call last):
  File "run_test.py", line 486, in <module>
    main()
  File "run_test.py", line 479, in main
    raise RuntimeError(message)
RuntimeError: test_jit_fuser failed!

fritzo · 2020-02-20T18:10:54Z

@pytestbot merge this please

facebook-github-bot

@ezyang is landing this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

…stream

facebook-github-bot

@ezyang is landing this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

facebook-github-bot · 2020-02-26T18:14:40Z

@ezyang merged this pull request in 24659d2.

Summary: Third try of #33177 😄 Pull Request resolved: #33418 Differential Revision: D20069683 Pulled By: ezyang fbshipit-source-id: f58e45e91b672bfde2e41a4480215ba4c613f9de

Summary: Third try of pytorch#33177 😄 Pull Request resolved: pytorch#33418 Differential Revision: D20069683 Pulled By: ezyang fbshipit-source-id: f58e45e91b672bfde2e41a4480215ba4c613f9de

driazati · 2020-03-13T02:01:46Z

torch/distributions/von_mises.py

+    return result
+
+
+@torch.jit.script


I didn't dig too much into the rest of this PR, but would be it OK to remove the @torch.jit.script here?

If this function is used in some other TorchScript method/module it will automatically be compiled, and having a @torch.jit.script decorator inside the core PyTorch library starts up the jit compiler anytime someone does import torch which does some expensive initialization (so it impacts all users of PyTorch instead of only those using the jit)

Yes, please go for it. Actually, I didn't realize we aren't supposed to not put torch.jit.script decorators in the core code; we should add a lint rule for this.

The reason I recommended @torch.jit.script here was to handle the data-dependent control flow when jit tracing code involving this distribution. In Pyro we use this pattern of @torch.jit.scripting low level helpers that cannot be traced. Do you have a recommended workaround in PyTorch core code, such that we can support torch.jit.trace?

I had a PR that did pretty much what you want (do nothing, but if tracing then compile the function) in #25746, I could revive that as something like @torch.jit._lazy_script

@driazati sure, a @torch.jit._lazy_script would be great.

Some users maintain libraries of code that is largely trace-able but not script-able. However, some functions may need to be `@torch.jit.script`ed if they contain control flow so the tracer will use the compiler version. This however impacts library start up time as in #33418, so this PR adds a workaround in the form of a `@torch.jit._lazy_script_while_tracing` that will only initialize the compiler if the function is called while actually tracing.

Summary: Stacked PRs * #34938 - [jit] Remove stray `script` * **#34935 - [jit] Add lazy script decorator** Some users maintain libraries of code that is largely trace-able but not script-able. However, some functions may need to be `torch.jit.script`ed if they contain control flow so the tracer will use the compiler version. This however impacts library start up time as in #33418, so this PR adds a workaround in the form of a `torch.jit._lazy_script_while_tracing` that will only initialize the compiler if the function is called while actually tracing. Pull Request resolved: #34935 Pulled By: driazati Differential Revision: D20569778 fbshipit-source-id: d87c88c02b1abc86b283729ab8db94285d7d4853

pytorchbot added the open source label Feb 17, 2020

ahmadsalim added 4 commits February 17, 2020 15:09

Moved VonMises distribution with sampling upstream from Pyro.

e757fbf

Add tests for VonMises logprob and sample Signed-off-by: Ahmad Salim Al-Sibahi <ahmad@di.ku.dk>

Optimize sampler test by using scipy (so it runs faster) + fix warnings

61de630

Wrap sampler in torch.jit.script

0bf59fa

Fix linting issues and JIT compilation Fix proposal_r Fix issues with JIT not working with torch.Size

[ci all] Fix issue with JIT of _log_modified_bessel

8f51f4c

ahmadsalim force-pushed the feature/vonmises-upstream branch from bc22baf to 8f51f4c Compare February 17, 2020 14:09

yf225 requested review from fritzo and ezyang February 19, 2020 21:05

yf225 added module: distributions Related to torch.distributions triaged This issue has been looked at a team member, and triaged and prioritized into an appropriate module labels Feb 19, 2020

fritzo reviewed Feb 19, 2020

View reviewed changes

test/test_distributions.py Outdated Show resolved Hide resolved

torch/distributions/von_mises.py Outdated Show resolved Hide resolved

Fix issues mentioned by @fritzo

ca2bf8a

fritzo approved these changes Feb 20, 2020

View reviewed changes

facebook-github-bot reviewed Feb 24, 2020

View reviewed changes

Merge remote-tracking branch 'origin/master' into feature/vonmises-up…

a3dfc73

…stream

facebook-github-bot reviewed Feb 25, 2020

View reviewed changes

ezyang approved these changes Feb 25, 2020

View reviewed changes

facebook-github-bot closed this in 24659d2 Feb 26, 2020

facebook-github-bot added the merged label Feb 26, 2020

hczhu pushed a commit that referenced this pull request Feb 28, 2020

Feature/vonmises upstream (#33418)

ea8f4b0

Summary: Third try of #33177 😄 Pull Request resolved: #33418 Differential Revision: D20069683 Pulled By: ezyang fbshipit-source-id: f58e45e91b672bfde2e41a4480215ba4c613f9de

driazati reviewed Mar 13, 2020

View reviewed changes

ahmadsalim deleted the feature/vonmises-upstream branch March 13, 2020 09:59

ezyang mentioned this pull request Mar 13, 2020

PCA and SVD for low-rank matrices, LOBPCG for positive-defined generalized eigenvalue problem #29488

Closed

30 tasks

driazati mentioned this pull request Mar 17, 2020

[jit] Remove stray @script #34929

Closed

driazati mentioned this pull request Mar 18, 2020

[jit] Add lazy script decorator #34935

Closed

mruberry added the Merged label Oct 28, 2020

mpariente mentioned this pull request Nov 12, 2020

Improve the execution time of the shift transform? asteroid-team/torch-audiomentations#41

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Feature/vonmises upstream #33418

Feature/vonmises upstream #33418

ahmadsalim commented Feb 17, 2020 •

edited

dr-ci bot commented Feb 17, 2020 •

edited

fritzo left a comment

fritzo left a comment

fritzo commented Feb 20, 2020

facebook-github-bot left a comment

facebook-github-bot left a comment

facebook-github-bot commented Feb 26, 2020

driazati Mar 13, 2020

ezyang Mar 13, 2020

fritzo Mar 13, 2020 •

edited

driazati Mar 13, 2020

fritzo Mar 13, 2020

Feature/vonmises upstream #33418

Feature/vonmises upstream #33418

Conversation

ahmadsalim commented Feb 17, 2020 • edited

dr-ci bot commented Feb 17, 2020 • edited

💊 CircleCI build failures summary and remediations

Detailed failure analysis

🕵️ 1 new failure recognized by patterns

pytorch_linux_xenial_py3_6_gcc5_4_test (1/1)

fritzo left a comment

Choose a reason for hiding this comment

fritzo left a comment

Choose a reason for hiding this comment

fritzo commented Feb 20, 2020

facebook-github-bot left a comment

Choose a reason for hiding this comment

facebook-github-bot left a comment

Choose a reason for hiding this comment

facebook-github-bot commented Feb 26, 2020

driazati Mar 13, 2020

Choose a reason for hiding this comment

ezyang Mar 13, 2020

Choose a reason for hiding this comment

fritzo Mar 13, 2020 • edited

Choose a reason for hiding this comment

driazati Mar 13, 2020

Choose a reason for hiding this comment

fritzo Mar 13, 2020

Choose a reason for hiding this comment

ahmadsalim commented Feb 17, 2020 •

edited

dr-ci bot commented Feb 17, 2020 •

edited

fritzo Mar 13, 2020 •

edited