add XPU support for IPEXModel.from_pretrained

[faaany]

What does this PR do?

This PR adds XPU support for loading a Torchscript model using IPEXModel.from_pretrained. Below is a test example:

import torch 
from transformers import AutoTokenizer
from optimum.intel import IPEXModel

model_id = "faaany/bert-base-uncased-float32-traced"
model_id_tokenizer = "google-bert/bert-base-uncased"
model = IPEXModel.from_pretrained(model_id)
tokenizer = AutoTokenizer.from_pretrained(model_id_tokenizer)
inputs = tokenizer("Paris is the capital of France.", return_tensors="pt").to("xpu")
print("________________")
print(model.device)
with torch.no_grad():
    outputs = model(**inputs)
    embeddings = outputs[0][:, 0]
    print(embeddings)

Please note that faaany/bert-base-uncased-float32-traced is just a test model, which is traced from "google-bert/bert-base-uncased". Since I didn't load any tokenizer config, so I pass model_id_tokenizer to AutoTokenizer in the example above.

[faaany]

@echarlaix pls have a review. Thx!

[faaany]

@yao-matrix

[faaany]

Hi @IlyasMoutawwakil could you help review this PR? Thanks a lot!

[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]

To fix the code style test you can do the following :

pip install .[quality]
make style

[faaany]

Hi @echarlaix, there are some CI tests failing. Do you know what I can do to fix them? Thx!

[echarlaix]

LGTM, thanks for the addition @faaany !