Add ONNX support for InternLM2

[gmf14]

What does this PR do?

Adds support for ONNX export of InternLM2 model for text generation tasks (#2240).

Note: I checked the ONNX contribution guide and it seems I cannot find an updated version to add the tests, hence I haven't committed them yet. I would like to kindly ask what steps to take in order to add the tests correctly.

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?

Who can review?

@fxmarty, @echarlaix, @JingyaHuang, @michaelbenayoun

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[echarlaix]

LGTM! Thanks a lot for the addition @gmf14

[echarlaix]

Could you add a test using a tiny random model like katuni4ka/tiny-random-internlm2 ?
You can add it in respectively PYTORCH_EXPORT_MODELS_TINY and MODEL_NAME

[gmf14]

Hi @echarlaix! Thanks a lot for your answer. I've added the tests in both PYTORCH_EXPORT_MODELS_TINY and MODEL_NAME.

[IlyasMoutawwakil]

you will have to pass trust_remote_code=True for this model since huggingface/transformers#29667 was not merged

[gmf14]

I'm figuring out what the best way to pass the trust_remote_code = True parameter to main_export() is for the text_export.py and text_export_cli.py files, for instance. Are there any other models that require this to true? A shallow approach (later can be abstracted) would be to simply add e.g. here something like:

# Check for models that require remote code execution
if "internlm2" in model_name:
    trust_remote_code = True

Something similar would apply for ORTModelForCausalLM in test_modeling.py.

Thanks a lot for your feedback @echarlaix @IlyasMoutawwakil!

[IlyasMoutawwakil]

@gmf14 for now we don't support many (any ?) of remote code models (because they are usually hard to maintain).
let's simply use if model_type in ["internlm2"]: trust_remote_code = True in both export and modeling tests.