构建一个 PDF 摄取和问答系统

先决条件

本指南假定您熟悉以下概念

PDF 文件通常包含来自其他来源无法获得的关键非结构化数据。它们可能相当长，并且与纯文本文件不同，通常无法直接馈送到语言模型的提示中。

在本教程中，您将创建一个能够回答有关 PDF 文件问题的系统。更具体地说，您将使用文档加载器以 LLM 可用的格式加载文本，然后构建一个检索增强生成 (RAG) 管道来回答问题，包括来自源材料的引用。

本教程将略过一些在我们 RAG 教程中更深入介绍的概念，因此如果您还没有，您可能希望先浏览它们。

让我们开始吧！

加载文档

首先，您需要选择要加载的 PDF。我们将使用来自耐克年度公开 SEC 报告的文档。它有超过 100 页，其中包含一些关键数据与更长的解释性文本混合在一起。但是，您可以随意使用您选择的 PDF。

选择 PDF 文件后，下一步是将它加载到大型语言模型 (LLM) 更容易处理的格式中，因为 LLM 通常需要文本输入。LangChain 提供了一些用于此目的的内置文档加载器，你可以进行尝试。下面，我们将使用一个由pypdf 包驱动的加载器，它从文件路径读取数据。

%pip install -qU pypdf langchain_community

from langchain_community.document_loaders import PyPDFLoader

file_path = "../example_data/nke-10k-2023.pdf"
loader = PyPDFLoader(file_path)

docs = loader.load()

print(len(docs))

API 参考：PyPDFLoader

print(docs[0].page_content[0:100])
print(docs[0].metadata)

Table of Contents
UNITED STATES
SECURITIES AND EXCHANGE COMMISSION
Washington, D.C. 20549
FORM 10-K

{'source': '../example_data/nke-10k-2023.pdf', 'page': 0}

发生了什么？

加载器将指定路径的 PDF 文件读取到内存中。
然后，它使用 pypdf 包提取文本数据。
最后，它为 PDF 的每一页创建一个 LangChain 文档，其中包含页面的内容以及有关文本来源的一些元数据。

LangChain 还有许多其他文档加载器可用于其他数据源，或者你可以创建一个自定义文档加载器。

使用 RAG 进行问答

接下来，你将准备加载的文档以供以后检索。使用文本分割器，你将加载的文档分割成更小的文档，这些文档更容易放入 LLM 的上下文窗口中，然后将它们加载到向量存储中。然后，你可以从向量存储中创建一个检索器，用于我们的 RAG 链。

pip install -qU langchain-openai

import getpass
import os

os.environ["OPENAI_API_KEY"] = getpass.getpass()

from langchain_openai import ChatOpenAI

llm = ChatOpenAI(model="gpt-4o")

pip install -qU langchain-anthropic

import getpass
import os

os.environ["ANTHROPIC_API_KEY"] = getpass.getpass()

from langchain_anthropic import ChatAnthropic

llm = ChatAnthropic(model="claude-3-5-sonnet-20240620")

pip install -qU langchain-openai

import getpass
import os

os.environ["AZURE_OPENAI_API_KEY"] = getpass.getpass()

from langchain_openai import AzureChatOpenAI

llm = AzureChatOpenAI(
    azure_endpoint=os.environ["AZURE_OPENAI_ENDPOINT"],
    azure_deployment=os.environ["AZURE_OPENAI_DEPLOYMENT_NAME"],
    openai_api_version=os.environ["AZURE_OPENAI_API_VERSION"],
)

pip install -qU langchain-google-vertexai

import getpass
import os

os.environ["GOOGLE_API_KEY"] = getpass.getpass()

from langchain_google_vertexai import ChatVertexAI

llm = ChatVertexAI(model="gemini-1.5-flash")

pip install -qU langchain-cohere

import getpass
import os

os.environ["COHERE_API_KEY"] = getpass.getpass()

from langchain_cohere import ChatCohere

llm = ChatCohere(model="command-r-plus")

pip install -qU langchain-nvidia-ai-endpoints

import getpass
import os

os.environ["NVIDIA_API_KEY"] = getpass.getpass()

from langchain import ChatNVIDIA

llm = ChatNVIDIA(model="meta/llama3-70b-instruct")

pip install -qU langchain-fireworks

import getpass
import os

os.environ["FIREWORKS_API_KEY"] = getpass.getpass()

from langchain_fireworks import ChatFireworks

llm = ChatFireworks(model="accounts/fireworks/models/llama-v3p1-70b-instruct")

pip install -qU langchain-groq

import getpass
import os

os.environ["GROQ_API_KEY"] = getpass.getpass()

from langchain_groq import ChatGroq

llm = ChatGroq(model="llama3-8b-8192")

pip install -qU langchain-mistralai

import getpass
import os

os.environ["MISTRAL_API_KEY"] = getpass.getpass()

from langchain_mistralai import ChatMistralAI

llm = ChatMistralAI(model="mistral-large-latest")

pip install -qU langchain-openai

import getpass
import os

os.environ["TOGETHER_API_KEY"] = getpass.getpass()

from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    base_url="https://api.together.xyz/v1",
    api_key=os.environ["TOGETHER_API_KEY"],
    model="mistralai/Mixtral-8x7B-Instruct-v0.1",
)

%pip install langchain_openai

from langchain_core.vectorstores import InMemoryVectorStore
from langchain_openai import OpenAIEmbeddings
from langchain_text_splitters import RecursiveCharacterTextSplitter

text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
splits = text_splitter.split_documents(docs)
vectorstore = InMemoryVectorStore.from_documents(
    documents=splits, embedding=OpenAIEmbeddings()
)

retriever = vectorstore.as_retriever()

API 参考：InMemoryVectorStore | OpenAIEmbeddings | RecursiveCharacterTextSplitter

最后，你将使用一些内置的助手来构建最终的 rag_chain。

from langchain.chains import create_retrieval_chain
from langchain.chains.combine_documents import create_stuff_documents_chain
from langchain_core.prompts import ChatPromptTemplate

system_prompt = (
    "You are an assistant for question-answering tasks. "
    "Use the following pieces of retrieved context to answer "
    "the question. If you don't know the answer, say that you "
    "don't know. Use three sentences maximum and keep the "
    "answer concise."
    "\n\n"
    "{context}"
)

prompt = ChatPromptTemplate.from_messages(
    [
        ("system", system_prompt),
        ("human", "{input}"),
    ]
)


question_answer_chain = create_stuff_documents_chain(llm, prompt)
rag_chain = create_retrieval_chain(retriever, question_answer_chain)

results = rag_chain.invoke({"input": "What was Nike's revenue in 2023?"})

results

API 参考：create_retrieval_chain | create_stuff_documents_chain | ChatPromptTemplate

{'input': "What was Nike's revenue in 2023?",
 'context': [Document(page_content='Table of Contents\nFISCAL 2023 NIKE BRAND REVENUE HIGHLIGHTS\nThe following tables present NIKE Brand revenues disaggregated by reportable operating segment, distribution channel and major product line:\nFISCAL 2023 COMPARED TO FISCAL 2022\n•NIKE, Inc. Revenues were $51.2 billion in fiscal 2023, which increased 10% and 16% compared to fiscal 2022 on a reported and currency-neutral basis, respectively.\nThe increase was due to higher revenues in North America, Europe, Middle East & Africa ("EMEA"), APLA and Greater China, which contributed approximately 7, 6,\n2 and 1 percentage points to NIKE, Inc. Revenues, respectively.\n•NIKE Brand revenues, which represented over 90% of NIKE, Inc. Revenues, increased 10% and 16% on a reported and currency-neutral basis, respectively. This\nincrease was primarily due to higher revenues in Men\'s, the Jordan Brand, Women\'s and Kids\' which grew 17%, 35%,11% and 10%, respectively, on a wholesale\nequivalent basis.', metadata={'page': 35, 'source': '../example_data/nke-10k-2023.pdf'}),
  Document(page_content='Enterprise Resource Planning Platform, data and analytics, demand sensing, insight gathering, and other areas to create an end-to-end technology foundation, which we\nbelieve will further accelerate our digital transformation. We believe this unified approach will accelerate growth and unlock more efficiency for our business, while driving\nspeed and responsiveness as we serve consumers globally.\nFINANCIAL HIGHLIGHTS\n•In fiscal 2023, NIKE, Inc. achieved record Revenues of $51.2 billion, which increased 10% and 16% on a reported and currency-neutral basis, respectively\n•NIKE Direct revenues grew 14% from $18.7 billion in fiscal 2022 to $21.3 billion in fiscal 2023, and represented approximately 44% of total NIKE Brand revenues for\nfiscal 2023\n•Gross margin for the fiscal year decreased 250 basis points to 43.5% primarily driven by higher product costs, higher markdowns and unfavorable changes in foreign\ncurrency exchange rates, partially offset by strategic pricing actions', metadata={'page': 30, 'source': '../example_data/nke-10k-2023.pdf'}),
  Document(page_content="Table of Contents\nNORTH AMERICA\n(Dollars in millions) FISCAL 2023FISCAL 2022 % CHANGE% CHANGE\nEXCLUDING\nCURRENCY\nCHANGESFISCAL 2021 % CHANGE% CHANGE\nEXCLUDING\nCURRENCY\nCHANGES\nRevenues by:\nFootwear $ 14,897 $ 12,228 22 % 22 %$ 11,644 5 % 5 %\nApparel 5,947 5,492 8 % 9 % 5,028 9 % 9 %\nEquipment 764 633 21 % 21 % 507 25 % 25 %\nTOTAL REVENUES $ 21,608 $ 18,353 18 % 18 %$ 17,179 7 % 7 %\nRevenues by:    \nSales to Wholesale Customers $ 11,273 $ 9,621 17 % 18 %$ 10,186 -6 % -6 %\nSales through NIKE Direct 10,335 8,732 18 % 18 % 6,993 25 % 25 %\nTOTAL REVENUES $ 21,608 $ 18,353 18 % 18 %$ 17,179 7 % 7 %\nEARNINGS BEFORE INTEREST AND TAXES $ 5,454 $ 5,114 7 % $ 5,089 0 %\nFISCAL 2023 COMPARED TO FISCAL 2022\n•North America revenues increased 18% on a currency-neutral basis, primarily due to higher revenues in Men's and the Jordan Brand. NIKE Direct revenues\nincreased 18%, driven by strong digital sales growth of 23%, comparable store sales growth of 9% and the addition of new stores.", metadata={'page': 39, 'source': '../example_data/nke-10k-2023.pdf'}),
  Document(page_content="Table of Contents\nEUROPE, MIDDLE EAST & AFRICA\n(Dollars in millions) FISCAL 2023FISCAL 2022 % CHANGE% CHANGE\nEXCLUDING\nCURRENCY\nCHANGESFISCAL 2021 % CHANGE% CHANGE\nEXCLUDING\nCURRENCY\nCHANGES\nRevenues by:\nFootwear $ 8,260 $ 7,388 12 % 25 %$ 6,970 6 % 9 %\nApparel 4,566 4,527 1 % 14 % 3,996 13 % 16 %\nEquipment 592 564 5 % 18 % 490 15 % 17 %\nTOTAL REVENUES $ 13,418 $ 12,479 8 % 21 %$ 11,456 9 % 12 %\nRevenues by:    \nSales to Wholesale Customers $ 8,522 $ 8,377 2 % 15 %$ 7,812 7 % 10 %\nSales through NIKE Direct 4,896 4,102 19 % 33 % 3,644 13 % 15 %\nTOTAL REVENUES $ 13,418 $ 12,479 8 % 21 %$ 11,456 9 % 12 %\nEARNINGS BEFORE INTEREST AND TAXES $ 3,531 $ 3,293 7 % $ 2,435 35 % \nFISCAL 2023 COMPARED TO FISCAL 2022\n•EMEA revenues increased 21% on a currency-neutral basis, due to higher revenues in Men's, the Jordan Brand, Women's and Kids'. NIKE Direct revenues\nincreased 33%, driven primarily by strong digital sales growth of 43% and comparable store sales growth of 22%.", metadata={'page': 40, 'source': '../example_data/nke-10k-2023.pdf'})],
 'answer': 'According to the financial highlights, Nike, Inc. achieved record revenues of $51.2 billion in fiscal 2023, which increased 10% on a reported basis and 16% on a currency-neutral basis compared to fiscal 2022.'}

你可以看到，你在结果字典的 answer 键中获得了最终答案，以及 LLM 用于生成答案的 context。

进一步检查 context 下的值，你可以看到它们是包含摄取页面内容一部分的文档。有用的是，这些文档还保留了最初加载它们时的原始元数据。

print(results["context"][0].page_content)

Table of Contents
FISCAL 2023 NIKE BRAND REVENUE HIGHLIGHTS
The following tables present NIKE Brand revenues disaggregated by reportable operating segment, distribution channel and major product line:
FISCAL 2023 COMPARED TO FISCAL 2022
•NIKE, Inc. Revenues were $51.2 billion in fiscal 2023, which increased 10% and 16% compared to fiscal 2022 on a reported and currency-neutral basis, respectively.
The increase was due to higher revenues in North America, Europe, Middle East & Africa ("EMEA"), APLA and Greater China, which contributed approximately 7, 6,
2 and 1 percentage points to NIKE, Inc. Revenues, respectively.
•NIKE Brand revenues, which represented over 90% of NIKE, Inc. Revenues, increased 10% and 16% on a reported and currency-neutral basis, respectively. This
increase was primarily due to higher revenues in Men's, the Jordan Brand, Women's and Kids' which grew 17%, 35%,11% and 10%, respectively, on a wholesale
equivalent basis.

print(results["context"][0].metadata)

{'page': 35, 'source': '../example_data/nke-10k-2023.pdf'}

这个特定的部分来自原始 PDF 的第 35 页。你可以使用此数据来显示答案来自 PDF 中的哪一页，从而使用户能够快速验证答案是否基于源材料。

信息

要深入了解 RAG，请查看这个更专注的教程或我们的操作指南。

下一步

你现在已经了解了如何使用文档加载器从 PDF 文件加载文档，以及可以用来准备加载的数据以进行 RAG 的一些技巧。

有关文档加载器的更多信息，你可以查看

有关 RAG 的更多信息，请查看

构建一个 PDF 摄取和问答系统

加载文档

使用 RAG 进行问答

下一步

此页面是否有帮助？

你也可以留下详细的反馈在 GitHub 上.

加载文档​

使用 RAG 进行问答​

下一步​

此页面是否有帮助？

你也可以留下详细的反馈 在 GitHub 上.

加载文档

使用 RAG 进行问答

下一步

你也可以留下详细的反馈在 GitHub 上.