Allow loading local models in a subfolder (with config in root)

[tomaarsen]

What does this PR do?

• Allow loading local models in a subfolder, also when the configuration is in root

Note: This PR depends on huggingface/optimum#2044 to be merged & released, and would require increasing the optimum version in the dependencies. I'm not sure how the optimum vs optimum-... dependencies are handled, so I'm not sure if this is a viable PR.

You can consider this a sister-PR to huggingface/optimum#2044 - the changes were also inspired from it.

What didn't work before?

import tempfile
from pathlib import Path

from huggingface_hub import HfApi
from optimum.intel import OVModelForFeatureExtraction

model_id = "sentence-transformers-testing/stsb-bert-tiny-openvino"

# hub model
ov_model = OVModelForFeatureExtraction.from_pretrained(model_id, subfolder="openvino", file_name="openvino_model.xml")
print("Hub model:", ov_model)

# local model
api = HfApi()
with tempfile.TemporaryDirectory() as tmpdirname:
    local_dir = Path(tmpdirname) / "model"
    api.snapshot_download(repo_id=model_id, local_dir=local_dir)

    ov_model = OVModelForFeatureExtraction.from_pretrained(local_dir, subfolder="openvino", file_name="openvino_model.xml")
    print("Remote model:", ov_model)

The second model from this snippet would not be loaded, because the subfolder was not being inserted between the model path and the filename. huggingface/optimum#2044 fixes the same issue but for ONNX.

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?

cc @echarlaix

• Tom Aarsen

[echarlaix]

LGTM, thanks @tomaarsen