Hugging Face integrations

This page covers all LangChain integrations with Hugging Face Hub and libraries like transformers, sentence transformers, and datasets.

Chat models

ChatHuggingFace

We can use the Hugging Face LLM classes or directly use the ChatHuggingFace class.

See a usage example.

from langchain_huggingface import ChatHuggingFace

LLMs

HuggingFaceEndpoint

We can use the HuggingFaceEndpoint class to run open source models via serverless Inference Providers or via dedicated Inference Endpoints.

See a usage example.

from langchain_huggingface import HuggingFaceEndpoint

HuggingFacePipeline

We can use the HuggingFacePipeline class to run open source models locally.

See a usage example.

from langchain_huggingface import HuggingFacePipeline

Embedding models

HuggingFaceEmbeddings

We can use the HuggingFaceEmbeddings class to run open source embedding models locally.

See a usage example.

from langchain_huggingface import HuggingFaceEmbeddings

HuggingFaceEndpointEmbeddings

We can use the HuggingFaceEndpointEmbeddings class to run open source embedding models via a dedicated Inference Endpoint.

See a usage example.

from langchain_huggingface import HuggingFaceEndpointEmbeddings

Text Embeddings Inference (TEI)

For self-hosted production serving of Sentence Transformers models, Hugging Face publishes Text Embeddings Inference, a dedicated inference server with batching and GPU support. TEI exposes an OpenAI-compatible API, so point LangChain at a TEI deployment via OpenAIEmbeddings. See the dedicated TEI integration guide.

BGE embedding models

BGE models on Hugging Face are a strong open-source embedding family from the Beijing Academy of Artificial Intelligence (BAAI).

BGE models are Sentence Transformers models, so use HuggingFaceEmbeddings with encode_kwargs={"normalize_embeddings": True}. See a usage example.