re-enable decoder sequence classification

[dwyatte]

What does this PR do?

Re-enables sequence classification for decoder-based models. This requires a user to pass in --pad_token_id which required a change to the tests

I also removed a default pad token override for GPT-2 which seemed like a bad idea (sets it to 0, which corresponds to !) -- let me know if this is problematic

This passes a fair number of tests in tests/exporters/onnx/ for me, but I deselected some so we may need to iterate here

Fixes #1527

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?

[fxmarty]

LGTM thank you for the fix!

[dwyatte]

Digging into some test failures

Failing on main:

FAILED onnxruntime/test_quantization.py::ORTDynamicQuantizationTest::test_dynamic_quantization_subgraphs - AttributeError: 'NoneType' object has no attribute 'HasField'

Failing on this branch:

FAILED exporters/onnx/test_exporters_onnx_cli.py::OnnxCLIExportTestCase::test_exporters_cli_pytorch_cpu_060_bloom_feature_extraction_with_past_default_hf_internal_testing_tiny_random_BloomModel - RuntimeError: dictionary changed size during iteration
FAILED exporters/onnx/test_exporters_onnx_cli.py::OnnxCLIExportTestCase::test_exporters_cli_pytorch_cpu_061_bloom_feature_extraction_with_past_no_postprocess_default_hf_internal_testing_tiny_random_BloomModel - RuntimeError: dictionary changed size during iteration
FAILED exporters/onnx/test_exporters_onnx_cli.py::OnnxCLIExportTestCase::test_exporters_cli_pytorch_cpu_no_dynamic_axes_07_gpt2_feature_extraction_with_past_default_hf_internal_testing_tiny_random_gpt2 - RuntimeError: dictionary changed size during iteration
FAILED exporters/onnx/test_exporters_onnx_cli.py::OnnxCLIExportTestCase::test_exporters_cli_pytorch_cpu_no_dynamic_axes_08_gpt2_feature_extraction_with_past_no_postprocess_default_hf_internal_testing_tiny_random_gpt2 - RuntimeError: dictionary changed size during iteration

but they pass locally for me. They appear to print out some warnings about tracing, wonder if my env is different than the CI env

========================================================================================================================================= warnings summary ==========================================================================================================================================
tests/exporters/onnx/test_exporters_onnx_cli.py::OnnxCLIExportTestCase::test_exporters_cli_pytorch_cpu_230_gpt2_feature_extraction_with_past_default_hf_internal_testing_tiny_random_gpt2
  /mnt/nfs/dwyatte/Development/cash-customer-support-ml/user-folders/dwyatte/optimum/venv/lib/python3.9/site-packages/torch/_utils.py:831: UserWarning: TypedStorage is deprecated. It will be removed in the future and UntypedStorage will be the only storage class. This should only matter to you if you are using storages directly.  To access UntypedStorage directly, use tensor.untyped_storage() instead of tensor.storage()
    return self.fget.__get__(instance, owner)()

tests/exporters/onnx/test_exporters_onnx_cli.py::OnnxCLIExportTestCase::test_exporters_cli_pytorch_cpu_230_gpt2_feature_extraction_with_past_default_hf_internal_testing_tiny_random_gpt2
tests/exporters/onnx/test_exporters_onnx_cli.py::OnnxCLIExportTestCase::test_exporters_cli_pytorch_cpu_231_gpt2_feature_extraction_with_past_no_postprocess_default_hf_internal_testing_tiny_random_gpt2
tests/exporters/onnx/test_exporters_onnx_cli.py::OnnxCLIExportTestCase::test_exporters_cli_pytorch_cpu_no_dynamic_axes_07_gpt2_feature_extraction_with_past_default_hf_internal_testing_tiny_random_gpt2
tests/exporters/onnx/test_exporters_onnx_cli.py::OnnxCLIExportTestCase::test_exporters_cli_pytorch_cpu_no_dynamic_axes_08_gpt2_feature_extraction_with_past_no_postprocess_default_hf_internal_testing_tiny_random_gpt2
  /mnt/nfs/dwyatte/Development/cash-customer-support-ml/user-folders/dwyatte/optimum/venv/lib/python3.9/site-packages/transformers/models/gpt2/modeling_gpt2.py:801: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
    if batch_size <= 0:

-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
====================================================================================================================== 614 passed, 2 skipped, 7 warnings in 163.98s (0:02:43) =======================================================================================================================

Results of optimum-cli env:

Copy-and-paste the text below in your GitHub issue:

- `optimum` version: 1.17.0.dev0
- `transformers` version: 4.37.2
- Platform: Linux-4.19.0-25-cloud-amd64-x86_64-with-glibc2.35
- Python version: 3.9.17
- Huggingface_hub version: 0.19.4
- PyTorch version (GPU?): 2.1.2+cu121 (cuda availabe: False)
- Tensorflow version (GPU?): not installed (cuda availabe: NA)

[fxmarty]

Thank you @dwyatte, these failing tests are fine as unrelated, and fixed in #1683 & microsoft/onnxruntime#19421

[dwyatte]

Great, just rebased so I think this should be ready to go now

[fxmarty]

Thank you!