如何在执行查询分析时处理多个查询

有时，查询分析技术可能允许生成多个查询。在这些情况下，我们需要记住运行所有查询，然后合并结果。我们将展示一个简单的示例（使用 mock 数据）来说明如何做到这一点。

设置

安装依赖项

%pip install -qU langchain langchain-community langchain-openai langchain-chroma

Note: you may need to restart the kernel to use updated packages.

设置环境变量

在此示例中，我们将使用 OpenAI：

import getpass
import os

if "OPENAI_API_KEY" not in os.environ:
    os.environ["OPENAI_API_KEY"] = getpass.getpass()

# Optional, uncomment to trace runs with LangSmith. Sign up here: https://smith.langchain.com.
# os.environ["LANGSMITH_TRACING"] = "true"
# os.environ["LANGSMITH_API_KEY"] = getpass.getpass()

创建索引

我们将在虚假信息上创建一个 vectorstore。

from langchain_chroma import Chroma
from langchain_openai import OpenAIEmbeddings
from langchain_text_splitters import RecursiveCharacterTextSplitter

texts = ["Harrison worked at Kensho", "Ankush worked at Facebook"]
embeddings = OpenAIEmbeddings(model="text-embedding-3-small")
vectorstore = Chroma.from_texts(
    texts,
    embeddings,
)
retriever = vectorstore.as_retriever(search_kwargs={"k": 1})

API 参考：OpenAI 嵌入 | 递归角色文本拆分器

查询分析

我们将使用函数调用来构建输出。我们将让它返回多个查询。

from typing import List, Optional

from pydantic import BaseModel, Field


class Search(BaseModel):
    """Search over a database of job records."""

    queries: List[str] = Field(
        ...,
        description="Distinct queries to search for",
    )

from langchain_core.output_parsers.openai_tools import PydanticToolsParser
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.runnables import RunnablePassthrough
from langchain_openai import ChatOpenAI

output_parser = PydanticToolsParser(tools=[Search])

system = """You have the ability to issue search queries to get information to help answer user information.

If you need to look up two distinct pieces of information, you are allowed to do that!"""
prompt = ChatPromptTemplate.from_messages(
    [
        ("system", system),
        ("human", "{question}"),
    ]
)
llm = ChatOpenAI(model="gpt-4o-mini", temperature=0)
structured_llm = llm.with_structured_output(Search)
query_analyzer = {"question": RunnablePassthrough()} | prompt | structured_llm

API 参考：PydanticToolsParser | ChatPromptTemplate 模板 | RunnablePassthrough | 聊天OpenAI

我们可以看到，这允许创建多个查询

query_analyzer.invoke("where did Harrison Work")

Search(queries=['Harrison Work', 'Harrison employment history'])

query_analyzer.invoke("where did Harrison and ankush Work")

Search(queries=['Harrison work history', 'Ankush work history'])

使用查询分析进行检索

那么我们如何将其包含在一个链中呢？使这变得容易得多的一件事是，如果我们异步调用我们的 retriever - 这将允许我们循环查询并且不会在响应时间上被阻塞。

from langchain_core.runnables import chain

API 参考：Chains

@chain
async def custom_chain(question):
    response = await query_analyzer.ainvoke(question)
    docs = []
    for query in response.queries:
        new_docs = await retriever.ainvoke(query)
        docs.extend(new_docs)
    # You probably want to think about reranking or deduplicating documents here
    # But that is a separate topic
    return docs

await custom_chain.ainvoke("where did Harrison Work")

[Document(page_content='Harrison worked at Kensho'),
 Document(page_content='Harrison worked at Kensho')]

await custom_chain.ainvoke("where did Harrison and ankush Work")

[Document(page_content='Harrison worked at Kensho'),
 Document(page_content='Ankush worked at Facebook')]