[Wav2Vec2] Fix torch srcipt

[patrickvonplaten]

What does this PR do?

Fixes Wav2Vec2 torch script slow tests

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

[patrickvonplaten]

It seems like torch.trace(...) doesn't like property function as it'll always call them. Since we've added the property as a private property in #23813, let's just go the simplest way and change it to a function.

This PR should fix:

tests/models/wav2vec2/test_modeling_wav2vec2.py::Wav2Vec2ModelTest::test_torchscript_output_attentions
(line 1180)  ValueError: <class 'transformers.models.wav2vec2.modeling_wav2vec2.Wav2Vec2ForCTC'> has no adapter layers. Make sure to define config.adapter_attn_dim.
tests/models/wav2vec2/test_modeling_wav2vec2.py::Wav2Vec2ModelTest::test_torchscript_output_hidden_state
(line 1180)  ValueError: <class 'transformers.models.wav2vec2.modeling_wav2vec2.Wav2Vec2ForCTC'> has no adapter layers. Make sure to define config.adapter_attn_dim.
tests/models/wav2vec2/test_modeling_wav2vec2.py::Wav2Vec2ModelTest::test_torchscript_simple
(line 1180)  ValueError: <class 'transformers.models.wav2vec2.modeling_wav2vec2.Wav2Vec2ForCTC'> has no adapter layers. Make sure to define config.adapter_attn_dim.
tests/models/wav2vec2/test_modeling_wav2vec2.py::Wav2Vec2RobustModelTest::test_torchscript_output_attentions
(line 1180)  ValueError: <class 'transformers.models.wav2vec2.modeling_wav2vec2.Wav2Vec2ForCTC'> has no adapter layers. Make sure to define config.adapter_attn_dim.
tests/models/wav2vec2/test_modeling_wav2vec2.py::Wav2Vec2RobustModelTest::test_torchscript_output_hidden_state
(line 1180)  ValueError: <class 'transformers.models.wav2vec2.modeling_wav2vec2.Wav2Vec2ForCTC'> has no adapter layers. Make sure to define config.adapter_attn_dim.
tests/models/wav2vec2/test_modeling_wav2vec2.py::Wav2Vec2RobustModelTest::test_torchscript_simple
(line 1180)  ValueError: <class 'transformers.models.wav2vec2.modeling_wav2vec2.Wav2Vec2ForCTC'> has no adapter layers. Make sure to define config.adapter_attn_dim.

of the slow tests, such as

RUN_SLOW=1 pytest tests/models/wav2vec2/test_modeling_wav2vec2.py::Wav2Vec2ModelTest::test_torchscript_simple

@sgugger @ydshieh

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[ydshieh]

the code in torch.jit.trace will call hasattr(mod, name). This function hasattr will actually call the method if it is a property. Let's why we have this failure. In general, not really a jit issue but the python thing. See this.

So good fore me, but would like to know how the core maintainers think.

[ydshieh]

One possible different way is to raise AttributeError instead of ValueError if we want to keep the propery.

[amyeroberts]

Change looks good to me - thanks for fixing!