[quant][eagermode] Move custom_module registration to prepare/convert_custom_config_dict

[jerryzh168]

Stack from ghstack:

• [quant][graphmode][fx] Support non_traceable_module/module_class #46298 [quant][graphmode][fx] Support non_traceable_module/module_class
• [quant][eagermode] Move custom_module registration to prepare/convert_custom_config_dict #46293 [quant][eagermode] Move custom_module registration to prepare/convert_custom_config_dict
• [quant] Remove prehook option #46292 [quant] Remove prehook option
• [quant][graphmode][fx] Move custom_module_class config to prepare/convert_custom_config_dict #46251 [quant][graphmode][fx] Move custom_module_class config to prepare/convert_custom_config_dict

Summary:

Test Plan:

Reviewers:

Subscribers:

Tasks:

Tags:

Differential Revision: D24290811

[dr-ci]

💊 CI failures summary and remediations

As of commit 8ce1bc6 (more details on the Dr. CI page):

---

• 5/5 failures possibly* introduced in this PR
  • 1/5 non-CircleCI failure(s)---

4 failures not recognized by patterns:

Job	Step	Action
pytorch_python_doc_build	Doc Build and Push	🔁 rerun
pytorch_bazel_test	Test	🔁 rerun
pytorch_libtorch_linux_xenial_cuda11_0_cudnn8_py3_gcc7_build	Build	🔁 rerun
pytorch_doc_test	Doc test	🔁 rerun

🚧 2 fixed upstream failures:

These were probably caused by upstream breakages that were already fixed.

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 https://github.com/pytorch/pytorch viable/strict
git rebase --onto FETCH_HEAD $(git merge-base origin/master HEAD)

If your commit is older than viable/strict:

git fetch https://github.com/pytorch/pytorch viable/strict
git rebase FETCH_HEAD

Check out the recency history of this "viable master" tracking branch.

• pytorch_libtorch_linux_xenial_cuda11_0_cudnn8_py3_gcc7_build on Oct 13 from 3:35pm to 6:58pm PDT (6 commits; 5c67cc7 - 1fcec6e)
  • 🔁 rerun
• pytorch_bazel_test on Oct 13 from 3:35pm to 6:58pm PDT (6 commits; 5c67cc7 - 1fcec6e)
  • 🔁 rerun

---

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 or post in the (internal) Dr. CI Users group.

See how this bot performed.

This comment has been revised 13 times.

[facebook-github-bot]

This pull request has been merged in 3ad797c.

[mruberry]

Unlanding. This broke the docs build. Relevant snippet:

Oct 14 20:44:57 /opt/conda/lib/python3.6/site-packages/torch/quantization/quantize.py:docstring of torch.quantization.prepare:20: WARNING: Block quote ends without a blank line; unexpected unindent.
Oct 14 20:44:57 /opt/conda/lib/python3.6/site-packages/torch/quantization/quantize.py:docstring of torch.quantization.convert:16: WARNING: Unexpected indentation.
Oct 14 20:44:57 /opt/conda/lib/python3.6/site-packages/torch/quantization/quantize.py:docstring of torch.quantization.convert:17: WARNING: Block quote ends without a blank line; unexpected unindent.

[jerryzh168]

Unlanding. This broke the docs build. Relevant snippet:

Oct 14 20:44:57 /opt/conda/lib/python3.6/site-packages/torch/quantization/quantize.py:docstring of torch.quantization.prepare:20: WARNING: Block quote ends without a blank line; unexpected unindent.
Oct 14 20:44:57 /opt/conda/lib/python3.6/site-packages/torch/quantization/quantize.py:docstring of torch.quantization.convert:16: WARNING: Unexpected indentation.
Oct 14 20:44:57 /opt/conda/lib/python3.6/site-packages/torch/quantization/quantize.py:docstring of torch.quantization.convert:17: WARNING: Block quote ends without a blank line; unexpected unindent.

oh, didn't notice that, I thought it was because of

Oct 14 01:08:03 /opt/conda/lib/python3.6/site-packages/torch/jit/_recursive.py:195: UserWarning: 'mods' was found in ScriptModule constants,  but it is a non-constant submodule. Consider removing it.
Oct 14 01:08:03   " but it is a non-constant {}. Consider removing it.".format(name, hint))
Oct 14 01:08:03 /opt/conda/lib/python3.6/site-packages/torch/jit/_trace.py:804: TracerWarning: Trace had nondeterministic nodes. Did you forget call .eval() on your model? Nodes:
Oct 14 01:08:03 	%13 : Float(4, strides=[1], requires_grad=0, device=cpu) = aten::rand(%8, %9, %10, %11, %12) # <doctest default[0]>:2:0
Oct 14 01:08:03 This may cause errors in trace checking. To disable trace checking, pass check_trace=False to torch.jit.trace()
Oct 14 01:08:03   _module_class,
Oct 14 01:08:03 /opt/conda/lib/python3.6/site-packages/torch/jit/_trace.py:804: TracerWarning: Output nr 1. of the traced function does not match the corresponding output of the Python function. Detailed error:
Oct 14 01:08:03 With rtol=1e-05 and atol=1e-05, found 4 element(s) (out of 12) whose difference(s) exceeded the margin of error (including 0 nan comparisons). The greatest difference was 0.8969195485115051 (0.0034521818161010742 vs. 0.9003717303276062), which occurred at index (0, 2).
Oct 14 01:08:03   _module_class,
Oct 14 01:08:03 /opt/conda/lib/python3.6/site-packages/torch/jit/_trace.py:804: TracerWarning: Trace had nondeterministic nodes. Did you forget call .eval() on your model? Nodes:
Oct 14 01:08:03 	%14 : Float(1, 4, strides=[4, 1], requires_grad=0, device=cpu) = aten::rand(%9, %10, %11, %12, %13) # <doctest default[0]>:2:0
Oct 14 01:08:03 This may cause errors in trace checking. To disable trace checking, pass check_trace=False to torch.jit.trace()
Oct 14 01:08:03   _module_class,
Oct 14 01:08:03 /opt/conda/lib/python3.6/site-packages/torch/jit/_trace.py:804: TracerWarning: Output nr 1. of the traced function does not match the corresponding output of the Python function. Detailed error:
Oct 14 01:08:03 With rtol=1e-05 and atol=1e-05, found 4 element(s) (out of 8) whose difference(s) exceeded the margin of error (including 0 nan comparisons). The greatest difference was 0.8881369233131409 (0.06778597831726074 vs. 0.9559229016304016), which occurred at index (0, 0).
Oct 14 01:08:03   _module_class,

can we make the error more explicit next time? why the quantization error does not appear in "Doctest summary"?

[mruberry]

Unlanding. This broke the docs build. Relevant snippet:

Oct 14 20:44:57 /opt/conda/lib/python3.6/site-packages/torch/quantization/quantize.py:docstring of torch.quantization.prepare:20: WARNING: Block quote ends without a blank line; unexpected unindent.
Oct 14 20:44:57 /opt/conda/lib/python3.6/site-packages/torch/quantization/quantize.py:docstring of torch.quantization.convert:16: WARNING: Unexpected indentation.
Oct 14 20:44:57 /opt/conda/lib/python3.6/site-packages/torch/quantization/quantize.py:docstring of torch.quantization.convert:17: WARNING: Block quote ends without a blank line; unexpected unindent.

oh, didn't notice that, I thought it was because of

Oct 14 01:08:03 /opt/conda/lib/python3.6/site-packages/torch/jit/_recursive.py:195: UserWarning: 'mods' was found in ScriptModule constants,  but it is a non-constant submodule. Consider removing it.
Oct 14 01:08:03   " but it is a non-constant {}. Consider removing it.".format(name, hint))
Oct 14 01:08:03 /opt/conda/lib/python3.6/site-packages/torch/jit/_trace.py:804: TracerWarning: Trace had nondeterministic nodes. Did you forget call .eval() on your model? Nodes:
Oct 14 01:08:03 	%13 : Float(4, strides=[1], requires_grad=0, device=cpu) = aten::rand(%8, %9, %10, %11, %12) # <doctest default[0]>:2:0
Oct 14 01:08:03 This may cause errors in trace checking. To disable trace checking, pass check_trace=False to torch.jit.trace()
Oct 14 01:08:03   _module_class,
Oct 14 01:08:03 /opt/conda/lib/python3.6/site-packages/torch/jit/_trace.py:804: TracerWarning: Output nr 1. of the traced function does not match the corresponding output of the Python function. Detailed error:
Oct 14 01:08:03 With rtol=1e-05 and atol=1e-05, found 4 element(s) (out of 12) whose difference(s) exceeded the margin of error (including 0 nan comparisons). The greatest difference was 0.8969195485115051 (0.0034521818161010742 vs. 0.9003717303276062), which occurred at index (0, 2).
Oct 14 01:08:03   _module_class,
Oct 14 01:08:03 /opt/conda/lib/python3.6/site-packages/torch/jit/_trace.py:804: TracerWarning: Trace had nondeterministic nodes. Did you forget call .eval() on your model? Nodes:
Oct 14 01:08:03 	%14 : Float(1, 4, strides=[4, 1], requires_grad=0, device=cpu) = aten::rand(%9, %10, %11, %12, %13) # <doctest default[0]>:2:0
Oct 14 01:08:03 This may cause errors in trace checking. To disable trace checking, pass check_trace=False to torch.jit.trace()
Oct 14 01:08:03   _module_class,
Oct 14 01:08:03 /opt/conda/lib/python3.6/site-packages/torch/jit/_trace.py:804: TracerWarning: Output nr 1. of the traced function does not match the corresponding output of the Python function. Detailed error:
Oct 14 01:08:03 With rtol=1e-05 and atol=1e-05, found 4 element(s) (out of 8) whose difference(s) exceeded the margin of error (including 0 nan comparisons). The greatest difference was 0.8881369233131409 (0.06778597831726074 vs. 0.9559229016304016), which occurred at index (0, 0).
Oct 14 01:08:03   _module_class,

can we make the error more explicit next time? why the quantization error does not appear in "Doctest summary"?

It IS confusing! @rgommers, let's consider improving this?

[rgommers]

What would you liked to have seen, something like?

NOTE: Sphinx warnings are turned into CI errors, this build is not clean!

right at the bottom of the pytorch_python_doc_build job, to make clear what is going on?

Current behaviour looks okay to me, but I realize that many devs aren't familiar with Sphinx and hence the build log may be hard to interpret.

Cc @mattip

[mattip]

Maybe this could be part of the logic in DrCI ?