Vald

Vald 是一個高度可擴展的分散式快速近似最近鄰 (ANN) 密集向量搜尋引擎。

本筆記本示範如何使用與 Vald 資料庫相關的功能。

若要執行本筆記本，您需要執行 Vald 叢集。請查看開始使用以取得更多資訊。

請參閱安裝指示。

%pip install --upgrade --quiet  vald-client-python langchain-community

基本範例

from langchain_community.document_loaders import TextLoader
from langchain_community.vectorstores import Vald
from langchain_huggingface import HuggingFaceEmbeddings
from langchain_text_splitters import CharacterTextSplitter

raw_documents = TextLoader("state_of_the_union.txt").load()
text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
documents = text_splitter.split_documents(raw_documents)
model_name = "sentence-transformers/all-mpnet-base-v2"
embeddings = HuggingFaceEmbeddings(model_name=model_name)
db = Vald.from_documents(documents, embeddings, host="localhost", port=8080)

API 參考：TextLoader | Vald | HuggingFaceEmbeddings | CharacterTextSplitter

query = "What did the president say about Ketanji Brown Jackson"
docs = db.similarity_search(query)
docs[0].page_content

依向量相似度搜尋

embedding_vector = embeddings.embed_query(query)
docs = db.similarity_search_by_vector(embedding_vector)
docs[0].page_content

依分數相似度搜尋

docs_and_scores = db.similarity_search_with_score(query)
docs_and_scores[0]

最大邊際相關性搜尋 (MMR)

除了在檢索器物件中使用相似度搜尋之外，您也可以使用 mmr 作為檢索器。

retriever = db.as_retriever(search_type="mmr")
retriever.invoke(query)

或直接使用 max_marginal_relevance_search

db.max_marginal_relevance_search(query, k=2, fetch_k=10)

使用安全連線的範例

為了執行本筆記本，必須使用安全連線執行 Vald 叢集。

以下是使用 Athenz 驗證的 Vald 叢集範例，其組態如下。

ingress(TLS) -> authorization-proxy(檢查 grpc metadata 中的 athenz-role-auth) -> vald-lb-gateway

import grpc

with open("test_root_cacert.crt", "rb") as root:
    credentials = grpc.ssl_channel_credentials(root_certificates=root.read())

# Refresh is required for server use
with open(".ztoken", "rb") as ztoken:
    token = ztoken.read().strip()

metadata = [(b"athenz-role-auth", token)]

from langchain_community.document_loaders import TextLoader
from langchain_community.vectorstores import Vald
from langchain_huggingface import HuggingFaceEmbeddings
from langchain_text_splitters import CharacterTextSplitter

raw_documents = TextLoader("state_of_the_union.txt").load()
text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
documents = text_splitter.split_documents(raw_documents)
model_name = "sentence-transformers/all-mpnet-base-v2"
embeddings = HuggingFaceEmbeddings(model_name=model_name)

db = Vald.from_documents(
    documents,
    embeddings,
    host="localhost",
    port=443,
    grpc_use_secure=True,
    grpc_credentials=credentials,
    grpc_metadata=metadata,
)

API 參考：TextLoader | Vald | HuggingFaceEmbeddings | CharacterTextSplitter

query = "What did the president say about Ketanji Brown Jackson"
docs = db.similarity_search(query, grpc_metadata=metadata)
docs[0].page_content

依向量相似度搜尋

embedding_vector = embeddings.embed_query(query)
docs = db.similarity_search_by_vector(embedding_vector, grpc_metadata=metadata)
docs[0].page_content

依分數相似度搜尋

docs_and_scores = db.similarity_search_with_score(query, grpc_metadata=metadata)
docs_and_scores[0]

最大邊際相關性搜尋 (MMR)

retriever = db.as_retriever(
    search_kwargs={"search_type": "mmr", "grpc_metadata": metadata}
)
retriever.invoke(query, grpc_metadata=metadata)

或

db.max_marginal_relevance_search(query, k=2, fetch_k=10, grpc_metadata=metadata)

相關內容

• 向量儲存庫概念指南
• 向量儲存庫操作指南