Missing model field in the huggingface_endpoint.py

[ENCODERS09]

Whatever model is being sent as a parameter it will never get called, only the default model will be called in the huggingface_endpoint.py because in the post call of the client and async_client the model is not being consumed which is why the default model is being called not the one passed

langchain/libs/partners/huggingface/langchain_huggingface/embeddings/huggingface_endpoint.py

Line 130 in b7e3e33

json={"inputs": texts, **_model_kwargs}, task=self.task

langchain/libs/partners/huggingface/langchain_huggingface/embeddings/huggingface_endpoint.py

Line 113 in b7e3e33

json={"inputs": texts, **_model_kwargs}, task=self.task

[keenborder786]

Nope the models is being passed during the initialization of InferenceClient

[ENCODERS09]

@keenborder786

https://github.com/huggingface/huggingface_hub/blob/f2e1889c66607528be17253f984fa2cd687d9b30/src/huggingface_hub/inference/_client.py#L266

If you see here, model is passed but the issue lies in the fact that, assigning the self.model attribute in the validate_environment method, but there is no ensuring that this assignment is persisted when we call the aembed_documents method.
The problem is In the validate_environment method, we are assigning the self.model attribute based on the self.repo_id attribute.
However, when we call the aembed_documents method, we are not re-validating the environment, so the self.model attribute might not be up-to-date.

So either we do this in aembed_documents() function

await self.validate_environment()
responses = await self.async_client.post(
    json={"inputs": texts, "parameters": _model_kwargs}, task=self.task
)

or we pass model=self.model

responses = await self.async_client.post(
    json={"inputs": texts, "parameters": _model_kwargs}, task=self.task, model=self.model
)