Fix emb model export and load with trfrs

[JingyaHuang]

What does this PR do?

Fixes #744

With the PR, we should be once again able to export embedding model via transformers library or sentence transformer library depending on the class called:

• With Transformers

import torch
from optimum.neuron import NeuronModelForFeatureExtraction
from transformers import AutoConfig, AutoTokenizer

compiler_args = {"auto_cast": "matmul", "auto_cast_type": "fp16"}
input_shapes = {"batch_size": 4, "sequence_length": 512}
model = NeuronModelForFeatureExtraction.from_pretrained(
    model_id="TaylorAI/bge-micro-v2", # BERT SMALL
    export=True,
    disable_neuron_cache=True,
    **compiler_args,
    **input_shapes,
)

• With Sentence Transformers

import torch
from optimum.neuron import NeuronModelForSentenceTransformers
from transformers import AutoConfig, AutoTokenizer

compiler_args = {"auto_cast": "matmul", "auto_cast_type": "fp16"}
input_shapes = {"batch_size": 4, "sequence_length": 512}
model = NeuronModelForSentenceTransformers.from_pretrained(
    model_id="TaylorAI/bge-micro-v2", # BERT SMALL
    export=True,
    disable_neuron_cache=True,
    **compiler_args,
    **input_shapes,
)

[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.

[dacorvo]

LGTM, Thanks ! We got hub errors on some of the CI jobs, but it is completely unrelated.