Skip to main content
Open In ColabOpen on GitHub

LLM Sherpa

本 Notebook 介绍了如何使用 LLM Sherpa 加载多种类型的文件。LLM Sherpa 支持不同的文件格式，包括 DOCX、PPTX、HTML、TXT 和 XML。

LLMSherpaFileLoader 使用 LayoutPDFReader，它是 LLMSherpa 库的一部分。该工具旨在解析 PDF 文件并保留其布局信息，这在使用大多数 PDF 转文本解析器时常常会丢失。

以下是 LayoutPDFReader 的一些主要功能：

  • 它可以识别和提取章节及子章节及其层级。
  • 它可以合并行以形成段落。
  • 它可以识别章节和段落之间的链接。
  • 它可以提取表格以及表格所在的章节。
  • 它可以识别并提取列表和嵌套列表。
  • 它可以连接跨页面的内容。
  • 它可以去除重复的页眉和页脚。
  • 它可以去除水印。

请查阅 llmsherpa 文档。

INFO: 此库在某些 PDF 文件上会失败，因此请谨慎使用。

# Install package
# !pip install --upgrade --quiet llmsherpa

LLMSherpaFileLoader​

LLMSherpaFileLoader 在底层定义了一些加载文件内容的策略：["sections", "chunks", "html", "text"]，设置 nlm-ingestor 以获取 llmsherpa_api_url 或使用默认值。

sections 策略：将文件解析为 sections 返回​

from langchain_community.document_loaders.llmsherpa import LLMSherpaFileLoader

loader = LLMSherpaFileLoader(
    file_path="https://arxiv.org/pdf/2402.14207.pdf",
    new_indent_parser=True,
    apply_ocr=True,
    strategy="sections",
    llmsherpa_api_url="http://localhost:5010/api/parseDocument?renderFormat=all",
)
docs = loader.load()
API Reference:LLMSherpaFileLoader
docs[1]
Document(page_content='Abstract\nWe study how to apply large language models to write grounded and organized long-form articles from scratch, with comparable breadth and depth to Wikipedia pages.\nThis underexplored problem poses new challenges at the pre-writing stage, including how to research the topic and prepare an outline prior to writing.\nWe propose STORM, a writing system for the Synthesis of Topic Outlines through\nReferences\nFull-length Article\nTopic\nOutline\n2022 Winter Olympics\nOpening Ceremony\nResearch via Question Asking\nRetrieval and Multi-perspective Question Asking.\nSTORM models the pre-writing stage by\nLLM\n(1) discovering diverse perspectives in researching the given topic, (2) simulating conversations where writers carrying different perspectives pose questions to a topic expert grounded on trusted Internet sources, (3) curating the collected information to create an outline.\nFor evaluation, we curate FreshWiki, a dataset of recent high-quality Wikipedia articles, and formulate outline assessments to evaluate the pre-writing stage.\nWe further gather feedback from experienced Wikipedia editors.\nCompared to articles generated by an outlinedriven retrieval-augmented baseline, more of STORM’s articles are deemed to be organized (by a 25% absolute increase) and broad in coverage (by 10%).\nThe expert feedback also helps identify new challenges for generating grounded long articles, such as source bias transfer and over-association of unrelated facts.\n1. Can you provide any information about the transportation arrangements for the opening ceremony?\nLLM\n2. Can you provide any information about the budget for the 2022 Winter Olympics opening ceremony?…\nLLM- Role1\nLLM- Role2\nLLM- Role1', metadata={'source': 'https://arxiv.org/pdf/2402.14207.pdf', 'section_number': 1, 'section_title': 'Abstract'})
len(docs)
79

分块策略：将文件解析为多个分块并返回​

from langchain_community.document_loaders.llmsherpa import LLMSherpaFileLoader

loader = LLMSherpaFileLoader(
    file_path="https://arxiv.org/pdf/2402.14207.pdf",
    new_indent_parser=True,
    apply_ocr=True,
    strategy="chunks",
    llmsherpa_api_url="http://localhost:5010/api/parseDocument?renderFormat=all",
)
docs = loader.load()
API Reference:LLMSherpaFileLoader
docs[1]
Document(page_content='Assisting in Writing Wikipedia-like Articles From Scratch with Large Language Models\nStanford University {shaoyj, yuchengj, tkanell, peterxu, okhattab}@stanford.edu lam@cs.stanford.edu', metadata={'source': 'https://arxiv.org/pdf/2402.14207.pdf', 'chunk_number': 1, 'chunk_type': 'para'})
len(docs)
306

HTML 策略：将文件作为单个 HTML 文档返回​

from langchain_community.document_loaders.llmsherpa import LLMSherpaFileLoader

loader = LLMSherpaFileLoader(
    file_path="https://arxiv.org/pdf/2402.14207.pdf",
    new_indent_parser=True,
    apply_ocr=True,
    strategy="html",
    llmsherpa_api_url="http://localhost:5010/api/parseDocument?renderFormat=all",
)
docs = loader.load()
API Reference:LLMSherpaFileLoader
docs[0].page_content[:400]
'<html><h1>Assisting in Writing Wikipedia-like Articles From Scratch with Large Language Models</h1><table><th><td colSpan=1>Yijia Shao</td><td colSpan=1>Yucheng Jiang</td><td colSpan=1>Theodore A. Kanell</td><td colSpan=1>Peter Xu</td></th><tr><td colSpan=1></td><td colSpan=1>Omar Khattab</td><td colSpan=1>Monica S. Lam</td><td colSpan=1></td></tr></table><p>Stanford University {shaoyj, yuchengj, '
len(docs)
1

文本策略：将文件作为单个文本文档返回​

from langchain_community.document_loaders.llmsherpa import LLMSherpaFileLoader

loader = LLMSherpaFileLoader(
    file_path="https://arxiv.org/pdf/2402.14207.pdf",
    new_indent_parser=True,
    apply_ocr=True,
    strategy="text",
    llmsherpa_api_url="http://localhost:5010/api/parseDocument?renderFormat=all",
)
docs = loader.load()
API Reference:LLMSherpaFileLoader
docs[0].page_content[:400]
'Assisting in Writing Wikipedia-like Articles From Scratch with Large Language Models\n | Yijia Shao | Yucheng Jiang | Theodore A. Kanell | Peter Xu\n | --- | --- | --- | ---\n |  | Omar Khattab | Monica S. Lam | \n\nStanford University {shaoyj, yuchengj, tkanell, peterxu, okhattab}@stanford.edu lam@cs.stanford.edu\nAbstract\nWe study how to apply large language models to write grounded and organized long'
len(docs)
1