Add BGE embeddings support (#8848)

- Description: [BGE-large](https://huggingface.co/BAAI/bge-large-en)
embeddings from BAAI are at the top of [MTEB
leaderboard](https://huggingface.co/spaces/mteb/leaderboard). Hence
adding support for it.
- Tag maintainer: @baskaryan
- Twitter handle: @ManabChetia3

---------

Co-authored-by: Harrison Chase <hw.chase.17@gmail.com>

docs/extras/integrations/text_embedding/bge_huggingface.ipynb

@@ -0,0 +1,84 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "id": "719619d3",
+   "metadata": {},
+   "source": [
+    "# BGE Hugging Face Embeddings\n",
+    "\n",
+    "This notebook shows how to use BGE Embeddings through Hugging Face"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 8,
+   "id": "f7a54279",
+   "metadata": {
+    "scrolled": true
+   },
+   "outputs": [],
+   "source": [
+    "# !pip install sentence_transformers"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "id": "9e1d5b6b",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from langchain.embeddings import HuggingFaceBgeEmbeddings\n",
+    "\n",
+    "model_name = \"BAAI/bge-small-en\"\n",
+    "model_kwargs = {'device': 'cpu'}\n",
+    "encode_kwargs = {'normalize_embeddings': False}\n",
+    "hf = HuggingFaceBgeEmbeddings(\n",
+    "    model_name=model_name,\n",
+    "    model_kwargs=model_kwargs,\n",
+    "    encode_kwargs=encode_kwargs\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 7,
+   "id": "e59d1a89",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "embedding = hf.embed_query(\"hi this is harrison\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "e596315f",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.10.1"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

libs/langchain/langchain/embeddings/__init__.py

@@ -31,6 +31,7 @@ from langchain.embeddings.fake import DeterministicFakeEmbedding, FakeEmbeddings
 from langchain.embeddings.google_palm import GooglePalmEmbeddings
 from langchain.embeddings.gpt4all import GPT4AllEmbeddings
 from langchain.embeddings.huggingface import (
+    HuggingFaceBgeEmbeddings,
     HuggingFaceEmbeddings,
     HuggingFaceInstructEmbeddings,
 )
@@ -97,6 +98,7 @@ __all__ = [
     "XinferenceEmbeddings",
     "LocalAIEmbeddings",
     "AwaEmbeddings",
+    "HuggingFaceBgeEmbeddings",
 ]
 
 

libs/langchain/langchain/embeddings/huggingface.py

@@ -6,10 +6,17 @@ from langchain.embeddings.base import Embeddings
 
 DEFAULT_MODEL_NAME = "sentence-transformers/all-mpnet-base-v2"
 DEFAULT_INSTRUCT_MODEL = "hkunlp/instructor-large"
+DEFAULT_BGE_MODEL = "BAAI/bge-large-en"
 DEFAULT_EMBED_INSTRUCTION = "Represent the document for retrieval: "
 DEFAULT_QUERY_INSTRUCTION = (
     "Represent the question for retrieving supporting documents: "
 )
+DEFAULT_EMBED_BGE_INSTRUCTION = (
+    "Represent this sentence for searching relevant passages: "
+)
+DEFAULT_QUERY_BGE_INSTRUCTION = (
+    "Represent this question for searching relevant passages: "
+)
 
 
 class HuggingFaceEmbeddings(BaseModel, Embeddings):
@@ -169,3 +176,86 @@ class HuggingFaceInstructEmbeddings(BaseModel, Embeddings):
         instruction_pair = [self.query_instruction, text]
         embedding = self.client.encode([instruction_pair], **self.encode_kwargs)[0]
         return embedding.tolist()
+
+
+class HuggingFaceBgeEmbeddings(BaseModel, Embeddings):
+    """HuggingFace BGE sentence_transformers embedding models.
+
+    To use, you should have the ``sentence_transformers`` python package installed.
+
+    Example:
+        .. code-block:: python
+
+            from langchain.embeddings import HuggingFaceBgeEmbeddings
+
+            model_name = "BAAI/bge-large-en"
+            model_kwargs = {'device': 'cpu'}
+            encode_kwargs = {'normalize_embeddings': False}
+            hf = HuggingFaceBgeEmbeddings(
+                model_name=model_name,
+                model_kwargs=model_kwargs,
+                encode_kwargs=encode_kwargs
+            )
+    """
+
+    client: Any  #: :meta private:
+    model_name: str = DEFAULT_BGE_MODEL
+    """Model name to use."""
+    cache_folder: Optional[str] = None
+    """Path to store models.
+    Can be also set by SENTENCE_TRANSFORMERS_HOME environment variable."""
+    model_kwargs: Dict[str, Any] = Field(default_factory=dict)
+    """Key word arguments to pass to the model."""
+    encode_kwargs: Dict[str, Any] = Field(default_factory=dict)
+    """Key word arguments to pass when calling the `encode` method of the model."""
+    embed_instruction: str = DEFAULT_EMBED_BGE_INSTRUCTION
+    """Instruction to use for embedding documents."""
+    query_instruction: str = DEFAULT_QUERY_BGE_INSTRUCTION
+    """Instruction to use for embedding query."""
+
+    def __init__(self, **kwargs: Any):
+        """Initialize the sentence_transformer."""
+        super().__init__(**kwargs)
+        try:
+            import sentence_transformers
+
+        except ImportError as exc:
+            raise ImportError(
+                "Could not import sentence_transformers python package. "
+                "Please install it with `pip install sentence_transformers`."
+            ) from exc
+
+        self.client = sentence_transformers.SentenceTransformer(
+            self.model_name, cache_folder=self.cache_folder, **self.model_kwargs
+        )
+
+    class Config:
+        """Configuration for this pydantic object."""
+
+        extra = Extra.forbid
+
+    def embed_documents(self, texts: List[str]) -> List[List[float]]:
+        """Compute doc embeddings using a HuggingFace transformer model.
+
+        Args:
+            texts: The list of texts to embed.
+
+        Returns:
+            List of embeddings, one for each text.
+        """
+        instruction_pairs = [[self.embed_instruction, text] for text in texts]
+        embeddings = self.client.encode(instruction_pairs, **self.encode_kwargs)
+        return embeddings.tolist()
+
+    def embed_query(self, text: str) -> List[float]:
+        """Compute query embeddings using a HuggingFace transformer model.
+
+        Args:
+            text: The text to embed.
+
+        Returns:
+            Embeddings for the text.
+        """
+        instruction_pair = [self.query_instruction, text]
+        embedding = self.client.encode([instruction_pair], **self.encode_kwargs)[0]
+        return embedding.tolist()