如何处理未生成查询的情况

有时，查询分析技术可能会允许生成任意数量的查询——包括不生成查询！在这种情况下，我们的整体链在决定是否调用检索器之前，需要检查查询分析的结果。

本例将使用模拟数据。

设置

安装依赖

%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

我们将创建一个基于虚假信息的向量存储。

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

texts = ["Harrison worked at Kensho"]
embeddings = OpenAIEmbeddings(model="text-embedding-3-small")
vectorstore = Chroma.from_texts(
    texts,
    embeddings,
)
retriever = vectorstore.as_retriever()

API Reference:Chroma | OpenAIEmbeddings | RecursiveCharacterTextSplitter

查询分析

我们将使用函数调用来构建输出。但我们会进行配置，使得 LLM 在决定不调用代表搜索查询的函数时也不需要这样做。然后，我们还将使用一个提示来进行查询分析，明确说明何时应该进行搜索，何时不应该。

from typing import Optional

from pydantic import BaseModel, Field


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

    query: str = Field(
        ...,
        description="Similarity search query applied to job record.",
    )

from langchain_core.prompts import ChatPromptTemplate
from langchain_core.runnables import RunnablePassthrough
from langchain_openai import ChatOpenAI

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

You do not NEED to look things up. If you don't need to, then just respond normally."""
prompt = ChatPromptTemplate.from_messages(
    [
        ("system", system),
        ("human", "{question}"),
    ]
)
llm = ChatOpenAI(model="gpt-4o-mini", temperature=0)
structured_llm = llm.bind_tools([Search])
query_analyzer = {"question": RunnablePassthrough()} | prompt | structured_llm

API Reference:ChatPromptTemplate | RunnablePassthrough | ChatOpenAI

我们可以看到，通过调用这个，我们会收到一条消息，该消息有时——但并非总是——返回一个工具调用。

query_analyzer.invoke("where did Harrison Work")

AIMessage(content='', additional_kwargs={'tool_calls': [{'id': 'call_korLZrh08PTRL94f4L7rFqdj', 'function': {'arguments': '{"query":"Harrison"}', 'name': 'Search'}, 'type': 'function'}], 'refusal': None}, response_metadata={'token_usage': {'completion_tokens': 14, 'prompt_tokens': 95, 'total_tokens': 109}, 'model_name': 'gpt-4o-mini-2024-07-18', 'system_fingerprint': 'fp_483d39d857', 'finish_reason': 'tool_calls', 'logprobs': None}, id='run-ea94d376-37bf-4f80-abe6-e3b42b767ea0-0', tool_calls=[{'name': 'Search', 'args': {'query': 'Harrison'}, 'id': 'call_korLZrh08PTRL94f4L7rFqdj', 'type': 'tool_call'}], usage_metadata={'input_tokens': 95, 'output_tokens': 14, 'total_tokens': 109})

query_analyzer.invoke("hi!")

AIMessage(content='Hello! How can I assist you today?', additional_kwargs={'refusal': None}, response_metadata={'token_usage': {'completion_tokens': 10, 'prompt_tokens': 93, 'total_tokens': 103}, 'model_name': 'gpt-4o-mini-2024-07-18', 'system_fingerprint': 'fp_483d39d857', 'finish_reason': 'stop', 'logprobs': None}, id='run-ebdfc44a-455a-4ca6-be85-84559886b1e1-0', usage_metadata={'input_tokens': 93, 'output_tokens': 10, 'total_tokens': 103})

通过查询分析进行检索

那么，我们如何将其包含在链中呢？让我们在下面的示例中看看。

from langchain_core.output_parsers.openai_tools import PydanticToolsParser
from langchain_core.runnables import chain

output_parser = PydanticToolsParser(tools=[Search])

API Reference:PydanticToolsParser | chain

@chain
def custom_chain(question):
    response = query_analyzer.invoke(question)
    if "tool_calls" in response.additional_kwargs:
        query = output_parser.invoke(response)
        docs = retriever.invoke(query[0].query)
        # Could add more logic - like another LLM call - here
        return docs
    else:
        return response

custom_chain.invoke("where did Harrison Work")

Number of requested results 4 is greater than number of elements in index 1, updating n_results = 1

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

custom_chain.invoke("hi!")

AIMessage(content='Hello! How can I assist you today?', additional_kwargs={'refusal': None}, response_metadata={'token_usage': {'completion_tokens': 10, 'prompt_tokens': 93, 'total_tokens': 103}, 'model_name': 'gpt-4o-mini-2024-07-18', 'system_fingerprint': 'fp_483d39d857', 'finish_reason': 'stop', 'logprobs': None}, id='run-e87f058d-30c0-4075-8a89-a01b982d557e-0', usage_metadata={'input_tokens': 93, 'output_tokens': 10, 'total_tokens': 103})