Milvus Hybrid Search Retriever

Overview

Milvus is an open-source vector database built to power embedding similarity search and AI applications. Milvus makes unstructured data search more accessible, and provides a consistent user experience regardless of the deployment environment.

This will help you getting started with the Milvus Hybrid Search retriever, which combines the strengths of both dense and sparse vector search. For detailed documentation of all MilvusCollectionHybridSearchRetriever features and configurations head to the API reference.

See also the Milvus Multi-Vector Search docs.

Integration details

Retriever	Self-host	Cloud offering	Package
MilvusCollectionHybridSearchRetriever	✅	❌	langchain_milvus

Setup

If you want to get automated tracing from individual queries, you can also set your LangSmith API key by uncommenting below:

# os.environ["LANGSMITH_API_KEY"] = getpass.getpass("Enter your LangSmith API key: ")
# os.environ["LANGSMITH_TRACING"] = "true"

Installation

This retriever lives in the langchain-milvus package. This guide requires the following dependencies:

%pip install --upgrade --quiet pymilvus[model] langchain-milvus langchain-openai

from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import PromptTemplate
from langchain_core.runnables import RunnablePassthrough
from langchain_milvus.retrievers import MilvusCollectionHybridSearchRetriever
from langchain_milvus.utils.sparse import BM25SparseEmbedding
from langchain_openai import ChatOpenAI, OpenAIEmbeddings
from pymilvus import (
    Collection,
    CollectionSchema,
    DataType,
    FieldSchema,
    WeightedRanker,
    connections,
)

Start the Milvus service

Please refer to the Milvus documentation to start the Milvus service.

After starting milvus, you need to specify your milvus connection URI.

CONNECTION_URI = "http://localhost:19530"

Prepare OpenAI API Key

Please refer to the OpenAI documentation to obtain your OpenAI API key, and set it as an environment variable.

export OPENAI_API_KEY=<your_api_key>

Prepare dense and sparse embedding functions

Let us fictionalize 10 fake descriptions of novels. In actual production, it may be a large amount of text data.

texts = [
    "In 'The Whispering Walls' by Ava Moreno, a young journalist named Sophia uncovers a decades-old conspiracy hidden within the crumbling walls of an ancient mansion, where the whispers of the past threaten to destroy her own sanity.",
    "In 'The Last Refuge' by Ethan Blackwood, a group of survivors must band together to escape a post-apocalyptic wasteland, where the last remnants of humanity cling to life in a desperate bid for survival.",
    "In 'The Memory Thief' by Lila Rose, a charismatic thief with the ability to steal and manipulate memories is hired by a mysterious client to pull off a daring heist, but soon finds themselves trapped in a web of deceit and betrayal.",
    "In 'The City of Echoes' by Julian Saint Clair, a brilliant detective must navigate a labyrinthine metropolis where time is currency, and the rich can live forever, but at a terrible cost to the poor.",
    "In 'The Starlight Serenade' by Ruby Flynn, a shy astronomer discovers a mysterious melody emanating from a distant star, which leads her on a journey to uncover the secrets of the universe and her own heart.",
    "In 'The Shadow Weaver' by Piper Redding, a young orphan discovers she has the ability to weave powerful illusions, but soon finds herself at the center of a deadly game of cat and mouse between rival factions vying for control of the mystical arts.",
    "In 'The Lost Expedition' by Caspian Grey, a team of explorers ventures into the heart of the Amazon rainforest in search of a lost city, but soon finds themselves hunted by a ruthless treasure hunter and the treacherous jungle itself.",
    "In 'The Clockwork Kingdom' by Augusta Wynter, a brilliant inventor discovers a hidden world of clockwork machines and ancient magic, where a rebellion is brewing against the tyrannical ruler of the land.",
    "In 'The Phantom Pilgrim' by Rowan Welles, a charismatic smuggler is hired by a mysterious organization to transport a valuable artifact across a war-torn continent, but soon finds themselves pursued by deadly assassins and rival factions.",
    "In 'The Dreamwalker's Journey' by Lyra Snow, a young dreamwalker discovers she has the ability to enter people's dreams, but soon finds herself trapped in a surreal world of nightmares and illusions, where the boundaries between reality and fantasy blur.",
]

We will use the OpenAI Embedding to generate dense vectors, and the BM25 algorithm to generate sparse vectors.

Initialize dense embedding function and get dimension

dense_embedding_func = OpenAIEmbeddings()
dense_dim = len(dense_embedding_func.embed_query(texts[1]))
dense_dim

1536

Initialize sparse embedding function.

Note that the output of sparse embedding is a set of sparse vectors, which represents the index and weight of the keywords of the input text.

sparse_embedding_func = BM25SparseEmbedding(corpus=texts)
sparse_embedding_func.embed_query(texts[1])

{0: 0.4270424944042204,
 21: 1.845826690498331,
 22: 1.845826690498331,
 23: 1.845826690498331,
 24: 1.845826690498331,
 25: 1.845826690498331,
 26: 1.845826690498331,
 27: 1.2237754316221157,
 28: 1.845826690498331,
 29: 1.845826690498331,
 30: 1.845826690498331,
 31: 1.845826690498331,
 32: 1.845826690498331,
 33: 1.845826690498331,
 34: 1.845826690498331,
 35: 1.845826690498331,
 36: 1.845826690498331,
 37: 1.845826690498331,
 38: 1.845826690498331,
 39: 1.845826690498331}

Create Milvus Collection and load data

Initialize connection URI and establish connection

connections.connect(uri=CONNECTION_URI)

Define field names and their data types

pk_field = "doc_id"
dense_field = "dense_vector"
sparse_field = "sparse_vector"
text_field = "text"
fields = [
    FieldSchema(
        name=pk_field,
        dtype=DataType.VARCHAR,
        is_primary=True,
        auto_id=True,
        max_length=100,
    ),
    FieldSchema(name=dense_field, dtype=DataType.FLOAT_VECTOR, dim=dense_dim),
    FieldSchema(name=sparse_field, dtype=DataType.SPARSE_FLOAT_VECTOR),
    FieldSchema(name=text_field, dtype=DataType.VARCHAR, max_length=65_535),
]

Create a collection with the defined schema

schema = CollectionSchema(fields=fields, enable_dynamic_field=False)
collection = Collection(
    name="IntroductionToTheNovels", schema=schema, consistency_level="Strong"
)

Define index for dense and sparse vectors

dense_index = {"index_type": "FLAT", "metric_type": "IP"}
collection.create_index("dense_vector", dense_index)
sparse_index = {"index_type": "SPARSE_INVERTED_INDEX", "metric_type": "IP"}
collection.create_index("sparse_vector", sparse_index)
collection.flush()

Insert entities into the collection and load the collection

entities = []
for text in texts:
    entity = {
        dense_field: dense_embedding_func.embed_documents([text])[0],
        sparse_field: sparse_embedding_func.embed_documents([text])[0],
        text_field: text,
    }
    entities.append(entity)
collection.insert(entities)
collection.load()

Instantiation

Now we can instantiate our retriever, defining search parameters for sparse and dense fields:

sparse_search_params = {"metric_type": "IP"}
dense_search_params = {"metric_type": "IP", "params": {}}
retriever = MilvusCollectionHybridSearchRetriever(
    collection=collection,
    rerank=WeightedRanker(0.5, 0.5),
    anns_fields=[dense_field, sparse_field],
    field_embeddings=[dense_embedding_func, sparse_embedding_func],
    field_search_params=[dense_search_params, sparse_search_params],
    top_k=3,
    text_field=text_field,
)

In the input parameters of this Retriever, we use a dense embedding and a sparse embedding to perform hybrid search on the two fields of this Collection, and use WeightedRanker for reranking. Finally, 3 top-K Documents will be returned.

Usage

retriever.invoke("What are the story about ventures?")

[Document(page_content="In 'The Lost Expedition' by Caspian Grey, a team of explorers ventures into the heart of the Amazon rainforest in search of a lost city, but soon finds themselves hunted by a ruthless treasure hunter and the treacherous jungle itself.", metadata={'doc_id': '449281835035545843'}),
 Document(page_content="In 'The Phantom Pilgrim' by Rowan Welles, a charismatic smuggler is hired by a mysterious organization to transport a valuable artifact across a war-torn continent, but soon finds themselves pursued by deadly assassins and rival factions.", metadata={'doc_id': '449281835035545845'}),
 Document(page_content="In 'The Dreamwalker's Journey' by Lyra Snow, a young dreamwalker discovers she has the ability to enter people's dreams, but soon finds herself trapped in a surreal world of nightmares and illusions, where the boundaries between reality and fantasy blur.", metadata={'doc_id': '449281835035545846'})]

Use within a chain

Initialize ChatOpenAI and define a prompt template

llm = ChatOpenAI()

PROMPT_TEMPLATE = """
Human: You are an AI assistant, and provides answers to questions by using fact based and statistical information when possible.
Use the following pieces of information to provide a concise answer to the question enclosed in <question> tags.

<context>
{context}
</context>

<question>
{question}
</question>

Assistant:"""

prompt = PromptTemplate(
    template=PROMPT_TEMPLATE, input_variables=["context", "question"]
)

Define a function for formatting documents

def format_docs(docs):
    return "\n\n".join(doc.page_content for doc in docs)

Define a chain using the retriever and other components

rag_chain = (
    {"context": retriever | format_docs, "question": RunnablePassthrough()}
    | prompt
    | llm
    | StrOutputParser()
)

Perform a query using the defined chain

rag_chain.invoke("What novels has Lila written and what are their contents?")

"Lila Rose has written 'The Memory Thief,' which follows a charismatic thief with the ability to steal and manipulate memories as they navigate a daring heist and a web of deceit and betrayal."

Drop the collection

collection.drop()

API reference

For detailed documentation of all MilvusCollectionHybridSearchRetriever features and configurations head to the API reference.