[Inference] Improve the support of sentence transformers

[JingyaHuang]

This PR aims

• To be aligned with recent changes in Optimum main repo: Proper sentence-transformers ONNX export support optimum#1589
• Offer better support for sentence transformers

It will solve reported issue: aws-neuron/aws-neuron-sdk#808

cc. @fxmarty

---

[Examples]

• Encoder

optimum-cli export neuron -m BAAI/bge-large-en-v1.5 --sequence_length 384 --batch_size 1 --task feature-extraction bge_emb/

• Clip

optimum-cli export neuron -m sentence-transformers/clip-ViT-B-32 --sequence_length 64 --batch_size 1 --num_channels 3 --height 64 --width 64 --task feature-extraction --library-name sentence_transformers --subfolder 0_CLIPModel clip_emb/

[Inference]

from transformers import AutoTokenizer
from optimum.neuron import NeuronModelForSenetenceTransformers

tokenizer = AutoTokenizer.from_pretrained("optimum/bge-base-en-v1.5-neuronx")
model = NeuronModelForSenetenceTransformers.from_pretrained("optimum/bge-base-en-v1.5-neuronx")

inputs = tokenizer("In the smouldering promise of the fall of Troy, a mythical world of gods and mortals rises from the ashes.", return_tensors="pt")

outputs = model(**inputs)
token_embeddings = outputs.token_embeddings
sentence_embedding = = outputs.sentence_embedding

[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, but do we have tests for this ?

[michaelbenayoun]

LGTM

[dacorvo]

LGTM, Thanks for this pull-request !

Note: "Abandon inf1": I sympathize ... 😉

[fxmarty]

@JingyaHuang as discussed we'll find a way to stop editing model_type for other libraries than transformers.

[JingyaHuang]

[Heads up!]
The inference API in this PR could be unstable as we are planning some refactoring for the root exporter in Optimum main.
cc. @fxmarty @echarlaix

[austinmw]

NeuronModelForSenetenceTransformers -> NeuronModelForSentenceTransformers

[JingyaHuang]

Good catch, thanks @austinmw !

(it's what happens when we copy a long string everywhere while having a typo... 🫣)

Fix will be here: #412