Add phi3 support in ONNX exporter

[JingyaHuang]

What does this PR do?

as per title

Fixes #1826

Example

(Needs transformers>=4.41.0)

• CLI

optimum-cli export onnx --model microsoft/Phi-3-mini-4k-instruct phi3_onnx/

• ORTModel

from transformers import AutoTokenizer

from optimum.onnxruntime import ORTModelForCausalLM


# [Export]
model_id = "microsoft/Phi-3-mini-4k-instruct"
model = ORTModelForCausalLM.from_pretrained(model_id, export=True)
model.save_pretrained("phi3_onnx/")

# [Inference]
model_id = "phi3_onnx/"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = ORTModelForCausalLM.from_pretrained(model_id)
inputs = tokenizer(["Hello, my dog is cute"], return_tensors="pt")
outputs = model.generate(**inputs, max_length=50)
print(tokenizer.batch_decode(outputs, skip_special_tokens=True))

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?

[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 @JingyaHuang !

[IlyasMoutawwakil]

LGTM ! thank you @JingyaHuang 🤗

[JingyaHuang]

Merging since failed CIs are irrelevant to the phi3 support.