Add MPNet ONNX export

[jplu]

What does this PR do?

This PR adds the possibility to export MPNet models into the ONNX format.

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]

Just for my knowledge, is opset 11 failing for this model? If you want, you can add a comment as to why opset 12 is required.

[jplu]

Here the error when export with opset 11:

An error occured with the error message: [ONNXRuntimeError] : 10 : INVALID_GRAPH : Load model from here/model.onnx failed:This is an invalid model. Type Error: Type 'tensor(int64)' of input parameter (/encoder/Add_1_output_0) of operator (Min) in node (/encoder/Min) is invalid..

Apparently the Min operator is not compliant with int64 tensors (as far as what I understand from this message)

[fxmarty]

Thank you for your contribution, looks good to me as long as the tests pass!

Edit: you may need to run make style

[HuggingFaceDocBuilderDev]

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

[jplu]

Looks like there an issue with the multiple_choice task:

Traceback (most recent call last):
  File "/opt/hostedtoolcache/Python/3.8.15/x64/lib/python3.8/runpy.py", line 194, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/opt/hostedtoolcache/Python/3.8.15/x64/lib/python3.8/runpy.py", line 87, in _run_code
    exec(code, run_globals)
  File "/opt/hostedtoolcache/Python/3.8.15/x64/lib/python3.8/site-packages/optimum/exporters/onnx/__main__.py", line 194, in <module>
    main()
  File "/opt/hostedtoolcache/Python/3.8.15/x64/lib/python3.8/site-packages/optimum/exporters/onnx/__main__.py", line 74, in main
    model = TasksManager.get_model_from_task(task, args.model, framework=args.framework, cache_dir=args.cache_dir)
  File "/opt/hostedtoolcache/Python/3.8.15/x64/lib/python3.8/site-packages/optimum/exporters/tasks.py", line 963, in get_model_from_task
    model = model_class.from_pretrained(model_name_or_path, **kwargs)
  File "/opt/hostedtoolcache/Python/3.8.15/x64/lib/python3.8/site-packages/transformers/models/auto/auto_factory.py", line 463, in from_pretrained
    return model_class.from_pretrained(
  File "/opt/hostedtoolcache/Python/3.8.15/x64/lib/python3.8/site-packages/transformers/modeling_utils.py", line 2379, in from_pretrained
    ) = cls._load_pretrained_model(
  File "/opt/hostedtoolcache/Python/3.8.15/x64/lib/python3.8/site-packages/transformers/modeling_utils.py", line 2695, in _load_pretrained_model
    raise RuntimeError(f"Error(s) in loading state_dict for {model.__class__.__name__}:\n\t{error_msg}")
RuntimeError: Error(s) in loading state_dict for MPNetForMultipleChoice:
	size mismatch for classifier.weight: copying a param with shape torch.Size([2, 64]) from checkpoint, the shape in current model is torch.Size([1, 64]).
	size mismatch for classifier.bias: copying a param with shape torch.Size([2]) from checkpoint, the shape in current model is torch.Size([1]).
	You may consider adding `ignore_mismatched_sizes=True` in the model `from_pretrained` method.

@fxmarty any idea how to fix this? Or should I remove this tasks from the task manager?

[fxmarty]

@jplu It's probably just an issue with hf-internal-testing/tiny-random-mpnet. Could you try using https://huggingface.co/hf-internal-testing/tiny-random-MPNetModel ?

You should be able to run the failing test specifying -k "test_exporters_cli_pytorch_273_mpnet_multiple_choice" in pytest

[jplu]

The test is skiped:

pytest tests/exporters/test_exporters_onnx_cli.py -k "test_exporters_cli_pytorch_273_mpnet_multiple_choice"
=================================================================================================================================================================================== test session starts ====================================================================================================================================================================================
platform linux -- Python 3.10.8, pytest-7.2.0, pluggy-1.0.0
rootdir: /home/jplu/dev/perso/optimum, configfile: setup.cfg
collected 413 items / 412 deselected / 1 selected

tests/exporters/test_exporters_onnx_cli.py s                                                                                                                                                                                                                                                                                                                                         [100%]

============================================================================================================================================================================ 1 skipped, 412 deselected in 1.41s ============================================================================================================================================================================

[fxmarty]

Hum that's weird. What if you just pass -k "mpnet"?

[jplu]

They are all skipped:

=================================================================================================================================================================================== test session starts ====================================================================================================================================================================================
platform linux -- Python 3.10.8, pytest-7.2.0, pluggy-1.0.0
rootdir: /home/jplu/dev/perso/optimum, configfile: setup.cfg
collected 413 items / 406 deselected / 7 selected

tests/exporters/test_exporters_onnx_cli.py sssssss                                                                                                                                                                                                                                                                                                                                   [100%]

============================================================================================================================================================================ 7 skipped, 406 deselected in 2.62s ============================================================================================================================================================================

[fxmarty]

I can run the mpnet tests locally on your branch with pytest tests/exporters/test_exporters_onnx_cli.py -k "mpnet" --exitfirst -s. Not sure why they are skipped for you?

(edit: the tests pass well with hf-internal-testing/tiny-random-MPNetModel)

[fxmarty]

It's all good, thanks for the addition!