Discrepancy between `client.sentence_similarity` input snippet & reality

[tomaarsen]

Hello!

Details

I've gotten an issue on one of my models (https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2/discussions/127) showing a discrepancy between the recommended usage snippet with the Inference Providers

import os
from huggingface_hub import InferenceClient

client = InferenceClient(
    provider="hf-inference",
    api_key=os.environ["HF_TOKEN"],
)

result = client.sentence_similarity(
    {
    "source_sentence": "That is a happy person",
    "sentences": [
        "That is a happy dog",
        "That is a very happy person",
        "Today is a sunny day"
    ]
},
    model="sentence-transformers/all-MiniLM-L6-v2",
)

and the actual code from huggingface_hub:

    def sentence_similarity(
        self, sentence: str, other_sentences: List[str], *, model: Optional[str] = None
    ) -> List[float]:

I suspect this is caused by this input:

huggingface.js/packages/tasks/src/snippets/inputs.ts

Lines 71 to 79 in e841a53

	const inputsSentenceSimilarity = () =>
	`{
	"source_sentence": "That is a happy person",
	"sentences": [
	"That is a happy dog",
	"That is a very happy person",
	"Today is a sunny day"
	]
	}`;

which should likely instead be

const inputsSentenceSimilarity = () => 
 	`{ 
     "sentence": "That is a happy person", 
     "other_sentences": [ 
         "That is a happy dog", 
         "That is a very happy person", 
         "Today is a sunny day" 
     ] 
 }`; 

or

const inputsSentenceSimilarity = () =>
	`sentence="That is a happy person",
    other_sentences=[
        "That is a happy dog",
        "That is a very happy person",
        "Today is a sunny day"
    ]`;

Depending on if the inputsSentenceSimilarity must be json parsable.
I'm afraid I don't currently have the bandwidth to make a PR for this.

cc @Wauplin @hanouticelina

• Tom Aarsen