LangChain - 文档转换

---

本文转载改编自：
https://python.langchain.com.cn/docs/modules/data_connection/document_transformers/

---

一、文档转换器 & 文本拆分器

一旦加载了文档，您通常会希望对其进行转换，以更好地适应您的应用程序。
最简单的例子是您可能希望将长文档拆分为更小的块，以适应您模型的上下文窗口。
LangChain提供了许多内置的文档转换器，使得拆分、合并、过滤和其他文档操作变得容易。

---

文本拆分器

当您想要处理大块文本时，有必要将文本拆分为块。
虽然听起来很简单，但这里存在许多 潜在的复杂性。

理想情况下，您希望将 语义相关的文本片段 保持在一起。

"语义相关"的含义可能取决于 文本的类型。本笔记本演示了几种做法。

---

在高层次上，文本拆分器的工作方式如下:

1. 将文本拆分为小的、语义上有意义的块（通常是句子）。
2. 将这些小块组合成较大的块，直到达到某个大小（由某个函数测量）。
3. 一旦达到该大小，将该块作为自己的文本片段，然后开始创建一个具有一定重叠的新文本块（以保持块之间的上下文）。

---

这意味着有两个不同的轴可以定制您的文本拆分器:

1. 文本如何拆分
2. 块大小如何测量

---

二、开始使用文本拆分器

默认推荐的文本分割器是 RecursiveCharacterTextSplitter。
该文本分割器接受一个字符列表。

它尝试根据第一个字符进行分割来创建块，但如果任何块太大，则继续移动到下一个字符，依此类推。

默认情况下，它尝试进行分割的字符是 ["\n\n", "\n", " ", ""]

---

除了控制可以进行分割的字符之外，您还可以控制一些其他事项：

• length_function：计算块长度的方法。默认只计算字符数，但通常在此处传递一个令牌计数器。
• chunk_size：块的最大大小（由长度函数测量）。
• chunk_overlap：块之间的最大重叠。保持一些连续性之间可能有一些重叠（例如使用滑动窗口）。
• add_start_index：是否在元数据中包含每个块在原始文档中的起始位置。

---

加载一段长文本

with open('../../state_of_the_union.txt') as f:state_of_the_union = f.read()

from langchain.text_splitter import RecursiveCharacterTextSplittertext_splitter = RecursiveCharacterTextSplitter(# Set a really small chunk size, just to show.chunk_size = 100,chunk_overlap  = 20,length_function = len,add_start_index = True,
)

---

texts = text_splitter.create_documents([state_of_the_union])
print(texts[0])
print(texts[1])

    page_content='Madam Speaker, Madam Vice President, our First Lady and Second Gentleman. Members of Congress and' metadata={'start_index': 0}page_content='of Congress and the Cabinet. Justices of the Supreme Court. My fellow Americans.' metadata={'start_index': 82}

---

三、按字符进行拆分

这是最简单的方法。它基于字符进行拆分（默认为"\n\n"），并通过字符数量来测量块的长度。

1. 文本如何被拆分: 按单个字符拆分。
2. 块大小如何被测量: 通过字符数量来测量。

---

# This is a long document we can split up.
with open('../../../state_of_the_union.txt') as f:state_of_the_union = f.read()

from langchain.text_splitter import CharacterTextSplitter
text_splitter = CharacterTextSplitter(        separator = "\n\n",chunk_size = 1000,chunk_overlap  = 200,length_function = len,
)

texts = text_splitter.create_documents([state_of_the_union])
print(texts[0])

    page_content='Madam Speaker, Madam Vice President, our First Lady and Second Gentleman. Members of Congress and the Cabinet. Justices of the Supreme Court. My fellow Americans. ...He met the Ukrainian people. \n\nFrom President Zelenskyy to every Ukrainian, their fearlessness, their courage, their determination, inspires the world.' lookup_str='' metadata={} lookup_index=0

---

如下示例，传递文档的元数据信息。注意，它是和文档一起拆分的。

metadatas = [{"document": 1}, {"document": 2}]
documents = text_splitter.create_documents([state_of_the_union, state_of_the_union], metadatas=metadatas)
print(documents[0])

    page_content='Madam Speaker, Madam Vice President, our First Lady and Second Gentleman. ...From President Zelenskyy to every Ukrainian, their fearlessness, their courage, their determination, inspires the world.' lookup_str='' metadata={'document': 1} lookup_index=0

---

text_splitter.split_text(state_of_the_union)[0]

    'Madam Speaker, Madam Vice President, our First Lady and Second Gentleman. Members of Congress and the Cabinet. Justices of the Supreme Court. My fellow Americans. ...From President Zelenskyy to every Ukrainian, their fearlessness, their courage, their determination, inspires the world.'

---

四、代码分割 (Split code)

CodeTextSplitter 允许您使用多种语言进行代码分割。

导入枚举 Language并指定语言。

from langchain.text_splitter import (RecursiveCharacterTextSplitter,Language,
)

Full list of support languages
[e.value for e in Language]

    ['cpp','go','java','js','php','proto','python','rst','ruby','rust','scala','swift','markdown','latex','html','sol',]

---

给定编程语言，你也可以看到 这个语言对应的 separators

RecursiveCharacterTextSplitter.get_separators_for_language(Language.PYTHON)

    ['\nclass ', '\ndef ', '\n\tdef ', '\n\n', '\n', ' ', '']

---

1、PythonTextSplitter

这里是使用 PythonTextSplitter 的示例

PYTHON_CODE = """
def hello_world():print("Hello, World!")# Call the function
hello_world()
"""
python_splitter = RecursiveCharacterTextSplitter.from_language(language=Language.PYTHON, chunk_size=50, chunk_overlap=0
)
python_docs = python_splitter.create_documents([PYTHON_CODE])
python_docs

    [Document(page_content='def hello_world():\n    print("Hello, World!")', metadata={}),Document(page_content='# Call the function\nhello_world()', metadata={})]

---

2、JS

这里是使用 JS 文本分割器的示例

JS_CODE = """
function helloWorld() {console.log("Hello, World!");
}// Call the function
helloWorld();
"""js_splitter = RecursiveCharacterTextSplitter.from_language(language=Language.JS, chunk_size=60, chunk_overlap=0
)
js_docs = js_splitter.create_documents([JS_CODE])
js_docs

    [Document(page_content='function helloWorld() {\n  console.log("Hello, World!");\n}', metadata={}),Document(page_content='// Call the function\nhelloWorld();', metadata={})]

---

3、Markdown

这里是使用 Markdown 文本分割器的示例

markdown_text = """# 🦜️🔗 LangChain⚡ Building applications with LLMs through composability ⚡## Quick Install```bashpip install langchain```As an open source project in a rapidly developing field, we are extremely open to contributions.
"""

md_splitter = RecursiveCharacterTextSplitter.from_language(language=Language.MARKDOWN, chunk_size=60, chunk_overlap=0
)
md_docs = md_splitter.create_documents([markdown_text])
md_docs

    [Document(page_content='# 🦜️🔗 LangChain', metadata={}),Document(page_content='⚡ Building applications with LLMs through composability ⚡', metadata={}),...Document(page_content='are extremely open to contributions.', metadata={})]

---

4、Latex

这里是使用 Latex 文本的示例

latex_text = """
\documentclass{article}\begin{document}\maketitle\section{Introduction}
Large language models (LLMs) are a type of machine learning model that can be trained on vast amounts of text data to generate human-like language. In recent years, LLMs have made significant advances in a variety of natural language processing tasks, including language translation, text generation, and sentiment analysis.\subsection{History of LLMs}
The earliest LLMs were developed in the 1980s and 1990s, but they were limited by the amount of data that could be processed and the computational power available at the time. In the past decade, however, advances in hardware and software have made it possible to train LLMs on massive datasets, leading to significant improvements in performance.\subsection{Applications of LLMs}
LLMs have many applications in industry, including chatbots, content creation, and virtual assistants. They can also be used in academia for research in linguistics, psychology, and computational linguistics.\end{document}
"""

latex_splitter = RecursiveCharacterTextSplitter.from_language(language=Language.MARKDOWN, chunk_size=60, chunk_overlap=0
)
latex_docs = latex_splitter.create_documents([latex_text])
latex_docs

[Document(page_content='\\documentclass{article}\n\n\x08egin{document}\n\n\\maketitle', metadata={}),Document(page_content='\\section{Introduction}', metadata={}),Document(page_content='Large language models (LLMs) are a type of machine learning', metadata={}),...Document(page_content='psychology, and computational linguistics.', metadata={}),Document(page_content='\\end{document}', metadata={})]

---

5、HTML

这里是使用 HTML 文本分割器的示例

html_text = """
<!DOCTYPE html>
<html><head><title>🦜️🔗 LangChain</title><style>body {font-family: Arial, sans-serif;}h1 {color: darkblue;}</style></head><body><div><h1>🦜️🔗 LangChain</h1><p>⚡ Building applications with LLMs through composability ⚡</p></div><div>As an open source project in a rapidly developing field, we are extremely open to contributions.</div></body>
</html>
"""

---

html_splitter = RecursiveCharacterTextSplitter.from_language(language=Language.MARKDOWN, chunk_size=60, chunk_overlap=0
)
html_docs = html_splitter.create_documents([html_text])
html_docs

---

    [Document(page_content='<!DOCTYPE html>\n<html>\n    <head>', metadata={}),Document(page_content='<title>🦜️🔗 LangChain</title>\n        <style>', metadata={}),Document(page_content='body {', metadata={}),Document(page_content='font-family: Arial, sans-serif;', metadata={}),Document(page_content='}\n            h1 {', metadata={}),Document(page_content='color: darkblue;\n            }', metadata={}),Document(page_content='</style>\n    </head>\n    <body>\n        <div>', metadata={}),Document(page_content='<h1>🦜️🔗 LangChain</h1>', metadata={}),Document(page_content='<p>⚡ Building applications with LLMs through', metadata={}),Document(page_content='composability ⚡</p>', metadata={}),Document(page_content='</div>\n        <div>', metadata={}),Document(page_content='As an open source project in a rapidly', metadata={}),Document(page_content='developing field, we are extremely open to contributions.', metadata={}),Document(page_content='</div>\n    </body>\n</html>', metadata={})]

---

6、Solidity

这里是使用 Solidity 文本分割器的示例

SOL_CODE = """
pragma solidity ^0.8.20;
contract HelloWorld {function add(uint a, uint b) pure public returns(uint) {return a + b;}
}
"""sol_splitter = RecursiveCharacterTextSplitter.from_language(language=Language.SOL, chunk_size=128, chunk_overlap=0
)
sol_docs = sol_splitter.create_documents([SOL_CODE])
sol_docs

[Document(page_content='pragma solidity ^0.8.20;', metadata={}),Document(page_content='contract HelloWorld {\n   function add(uint a, uint b) pure public returns(uint) {\n       return a + b;\n   }\n}', metadata={})
]

---

五、MarkdownHeaderTextSplitter

---

1、动机

许多聊天或问答应用程序在嵌入和向量存储之前，会先对输入文档进行分割成块。

Pinecone 的这些笔记提供了一些有用的提示：

当嵌入整个段落或文档时，嵌入过程会同时考虑整体上下文和文本中句子和短语之间的关系。这可能会得到更全面的向量表示，捕捉文本的更广泛的含义和主题。

正如上面所述，分块通常旨在将具有共同上下文的文本保持在一起。

在这种情况下，我们可能想要特别尊重文档本身的结构。

例如，一个 Markdown 文件的组织方式是通过标题。

在特定的标题组内创建分块是一个直观的想法。

为了解决这个挑战，我们可以使用 MarkdownHeaderTextSplitter。

它将按照指定的一组标题来分割一个 Markdown 文件。

例如，如果我们想要分割这个 Markdown：

md = '# Foo\n\n ## Bar\n\nHi this is Jim  \nHi this is Joe\n\n ## Baz\n\n Hi this is Molly' 

---

我们可以指定要分割的标题：

[("#", "Header 1"),("##", "Header 2")]

---

然后根据公共标题进行内容的分组或分割：

{'content': 'Hi this is Jim  \nHi this is Joe', 'metadata': {'Header 1': 'Foo', 'Header 2': 'Bar'}}
{'content': 'Hi this is Molly', 'metadata': {'Header 1': 'Foo', 'Header 2': 'Baz'}}

---

让我们来看一些下面的示例。

from langchain.text_splitter import MarkdownHeaderTextSplitter

markdown_document = "# Foo\n\n    ## Bar\n\nHi this is Jim\n\nHi this is Joe\n\n ### Boo \n\n Hi this is Lance \n\n ## Baz\n\n Hi this is Molly"headers_to_split_on = [("#", "Header 1"),("##", "Header 2"),("###", "Header 3"),
]markdown_splitter = MarkdownHeaderTextSplitter(headers_to_split_on=headers_to_split_on)
md_header_splits = markdown_splitter.split_text(markdown_document)
for split in md_header_splits:print(split)

{'content': 'Hi this is Jim  \nHi this is Joe', 'metadata': {'Header 1': 'Foo', 'Header 2': 'Bar'}}
{'content': 'Hi this is Lance', 'metadata': {'Header 1': 'Foo', 'Header 2': 'Bar', 'Header 3': 'Boo'}}
{'content': 'Hi this is Molly', 'metadata': {'Header 1': 'Foo', 'Header 2': 'Baz'}}

---

在每个 markdown 组中，我们可以应用我们需要的 text splitter。

markdown_document = "# Intro \n\n    ## History \n\n Markdown[9] is a lightweight markup language for creating formatted text using a plain-text editor. John Gruber created Markdown in 2004 as a markup language that is appealing to human readers in its source code form.[9] \n\n Markdown is widely used in blogging, instant messaging, online forums, collaborative software, documentation pages, and readme files. \n\n ## Rise and divergence \n\n As Markdown popularity grew rapidly, many Markdown implementations appeared, driven mostly by the need for \n\n additional features such as tables, footnotes, definition lists,[note 1] and Markdown inside HTML blocks. \n\n #### Standardization \n\n From 2012, a group of people, including Jeff Atwood and John MacFarlane, launched what Atwood characterised as a standardisation effort. \n\n ## Implementations \n\n Implementations of Markdown are available for over a dozen programming languages."headers_to_split_on = [("#", "Header 1"),("##", "Header 2"),
]# MD splits
markdown_splitter = MarkdownHeaderTextSplitter(headers_to_split_on=headers_to_split_on)
md_header_splits = markdown_splitter.split_text(markdown_document)# Char-level splits
from langchain.text_splitter import RecursiveCharacterTextSplitter
chunk_size = 10
chunk_overlap = 0
text_splitter = RecursiveCharacterTextSplitter(chunk_size=chunk_size, chunk_overlap=chunk_overlap)# Split within each header group
all_splits=[]
all_metadatas=[]    
for header_group in md_header_splits:_splits = text_splitter.split_text(header_group['content'])_metadatas = [header_group['metadata'] for _ in _splits]all_splits += _splitsall_metadatas += _metadatas

---

all_splits[0]
# -> 'Markdown[9'

all_metadatas[0]
# -> {'Header 1': 'Intro', 'Header 2': 'History'}

---

2、Use case

我们将 MarkdownHeaderTextSplitter 应用到 Notion page 作为测试。详情可见：https://rlancemartin.notion.site/Auto-Evaluation-of-Metadata-Filtering-18502448c85240828f33716740f9574b

这个页面使用 markdown 下载保存到本地。

# Load Notion database as a markdownfile file
from langchain.document_loaders import NotionDirectoryLoader
loader = NotionDirectoryLoader("../Notion_DB_Metadata")
docs = loader.load()
md_file=docs[0].page_content

# Let's create groups based on the section headers
headers_to_split_on = [("###", "Section"),
]
markdown_splitter = MarkdownHeaderTextSplitter(headers_to_split_on=headers_to_split_on)
md_header_splits = markdown_splitter.split_text(md_file)
md_header_splits[3]

{'content': 'We previously introduced [auto-evaluator](https://blog.langchain.dev/auto-evaluator-opportunities/), an open-source tool for grading LLM question-answer chains. Here, we extend auto-evaluator with a [lightweight Streamlit app](https://github.com/langchain-ai/auto-evaluator/tree/main/streamlit) that can connect to any existing Pinecone index. We add the ability to test metadata filtering using `SelfQueryRetriever` as well as some other approaches that we’ve found to be useful, as discussed below.  \n[ret_trim.mov](Auto-Evaluation%20of%20Metadata%20Filtering%2018502448c85240828f33716740f9574b/ret_trim.mov)','metadata': {'Section': 'Evaluation'}}

---

现在，我们将文本拆分到每个组中，并将该组作为元数据保存。

# Define our text splitter
from langchain.text_splitter import RecursiveCharacterTextSplitter
chunk_size = 500
chunk_overlap = 50
text_splitter = RecursiveCharacterTextSplitter(chunk_size=chunk_size, chunk_overlap=chunk_overlap)# Create splits within each header group
all_splits=[]
all_metadatas=[]
for header_group in md_header_splits:_splits = text_splitter.split_text(header_group['content'])_metadatas = [header_group['metadata'] for _ in _splits]all_splits += _splitsall_metadatas += _metadatas

all_splits[6]

'In these cases, semantic search will look for the concept `episode 53` in the chunks, but instead we simply want to filter the chunks for `episode 53` and then perform semantic search to extract those that best summarize the episode. Metadata filtering does this, so long as we 1) we have a metadata filter for episode number and 2) we can extract the value from the query (e.g., `54` or `252`) that we want to extract. The LangChain `SelfQueryRetriever` does the latter (see'

---

all_metadatas[6]

{'Section': 'Motivation'}

---

这使我们能够很好地执行 基于文档结构的 元数据过滤。

让我们先建一个向量库，把这一切结合起来。

! pip install chromadb

# Build vectorstore
from langchain.vectorstores import Chroma
from langchain.embeddings.openai import OpenAIEmbeddings
embeddings = OpenAIEmbeddings()
vectorstore = Chroma.from_texts(texts=all_splits,metadatas=all_metadatas,embedding=OpenAIEmbeddings())

---

我们创建一个 SelfQueryRetriever，可以根据我们定义的元数据进行筛选。

# Create retriever 
from langchain.llms import OpenAI
from langchain.retrievers.self_query.base import SelfQueryRetriever
from langchain.chains.query_constructor.base import AttributeInfo# Define our metadata
metadata_field_info = [AttributeInfo(name="Section",description="Headers of the markdown document that organize the ideas",type="string or list[string]",),
]
document_content_description = "Headers of the markdown document"# Define self query retriver
llm = OpenAI(temperature=0)
sq_retriever = SelfQueryRetriever.from_llm(llm, vectorstore, document_content_description, metadata_field_info, verbose=True)

---

然后我们可以从 文章的任意部分，获取 chunks。

# Test
question="Summarize the Introduction section of the document"
sq_retriever.get_relevant_documents(question)

query='Introduction' filter=Comparison(comparator=<Comparator.EQ: 'eq'>, attribute='Section', value='Introduction') limit=None[Document(page_content='![Untitled](Auto-Evaluation%20of%20Metadata%20Filtering%2018502448c85240828f33716740f9574b/Untitled.png)', metadata={'Section': 'Introduction'}),Document(page_content='Q+A systems often use a two-step approach: retrieve relevant text chunks and then synthesize them into an answer. ... Metadata filtering is an alternative approach that pre-filters chunks based on a user-defined criteria in a VectorDB using', metadata={'Section': 'Introduction'}),Document(page_content='on a user-defined criteria in a VectorDB using metadata tags prior to semantic search.', metadata={'Section': 'Introduction'})]

---

现在，我们可以创建清洗的文档结构的 聊天或Q+A 应用程序。

当然，没有特定元数据过滤的语义搜索，可能对这个简单的文档 相当有效。

但是，对于更复杂或更长的文档，保留文档结构 以进行 元数据过滤的能力 可能会有所帮助。

from langchain.chains import RetrievalQA
from langchain.chat_models import ChatOpenAI
llm = ChatOpenAI(model_name="gpt-3.5-turbo", temperature=0)
qa_chain = RetrievalQA.from_chain_type(llm,retriever=sq_retriever)
qa_chain.run(question)

query='Introduction' filter=Comparison(comparator=<Comparator.EQ: 'eq'>, attribute='Section', value='Introduction') limit=None'The document discusses different approaches to retrieve relevant text chunks and synthesize them into an answer in Q+A systems. 
...
The Retriever-Less option, which uses the Anthropic 100k context window model, is also mentioned as an alternative approach.'

question="Summarize the Testing section of the document"
qa_chain.run(question)

query='Testing' filter=Comparison(comparator=<Comparator.EQ: 'eq'>, attribute='Section', value='Testing') limit=None'The Testing section of the document describes how the performance of the SelfQueryRetriever was evaluated using various test cases. 
...
Additionally, the document mentions the use of the Kor library for structured data extraction to explicitly specify transformations that the auto-evaluator can use.'

---

六、递归按字符分割

这个文本分割器 是用于 通用文本的推荐分割器。它通过一个 字符列表进行参数化。
它会按 顺序 尝试使用这些字符进行分割，直到块的大小足够小。
默认列表是 ["\n\n", "\n", " ", ""]。

这样做的效果是尽可能地保持所有段落（然后是句子，然后是单词）在一起，因为它们通常是 在语义上相关的文本片段中 的最强关联部分。

1. 文本如何分割：按字符列表。
2. 块的大小如何衡量：按字符数。

This is a long document we can split up.
with open('../../../state_of_the_union.txt') as f:state_of_the_union = f.read()

---

from langchain.text_splitter import RecursiveCharacterTextSplitter
text_splitter = RecursiveCharacterTextSplitter(# Set a really small chunk size, just to show.chunk_size = 100,chunk_overlap  = 20,length_function = len,
)

texts = text_splitter.create_documents([state_of_the_union])
print(texts[0])
print(texts[1])

    page_content='Madam Speaker, Madam Vice President, our First Lady and Second Gentleman. Members of Congress and' lookup_str='' metadata={} lookup_index=0page_content='of Congress and the Cabinet. Justices of the Supreme Court. My fellow Americans.' lookup_str='' metadata={} lookup_index=0

---

text_splitter.split_text(state_of_the_union)[:2]

    ['Madam Speaker, Madam Vice President, our First Lady and Second Gentleman. Members of Congress and','of Congress and the Cabinet. Justices of the Supreme Court. My fellow Americans.']

---

七、按token 进行分割

语言模型有一个token限制。您不应超过token限制。
因此，当您将文本分割成块时，将token的数量进行计数是一个好主意。
有许多分词器可供使用。在计数文本中的token时，应使用 与语言模型 中使用的相同的分词器。

---

1、tiktoken

tiktoken 是由 OpenAI 创建的高速BPE分词器。

我们可以使用它来估计已使用的token。对于 OpenAI 模型，它可能更准确。

1. 文本的分割方式：通过传入的字符进行分割
2. 分块大小的衡量标准：使用 tiktoken 分词器计数

---

安装 tiktoken

!pip install tiktoken

# This is a long document we can split up.
with open("../../../state_of_the_union.txt") as f:state_of_the_union = f.read()
from langchain.text_splitter import CharacterTextSplitter

text_splitter = CharacterTextSplitter.from_tiktoken_encoder(chunk_size=100, chunk_overlap=0
)
texts = text_splitter.split_text(state_of_the_union)

---

texts[0]

Madam Speaker, Madam Vice President, our First Lady and Second Gentleman. Members of Congress and the Cabinet. Justices of the Supreme Court. My fellow Americans.  Last year COVID-19 kept us apart. This year we are finally together again. Tonight, we meet as Democrats Republicans and Independents. But most importantly as Americans. With a duty to one another to the American people to the Constitution.

---

也可以直接 load 一个 tiktoken splitter

from langchain.text_splitter import TokenTextSplittertext_splitter = TokenTextSplitter(chunk_size=10, chunk_overlap=0)texts = text_splitter.split_text(state_of_the_union)
print(texts[0])

---

2、spaCy

spaCy is an open-source software library for advanced natural language processing, written in the programming languages Python and Cython.

另一个替代NLTK 的是 spaCy tokenizer.

1. How the text is split: by spaCy tokenizer
2. How the chunk size is measured: by number of characters

---

!pip install spacy

# This is a long document we can split up.
with open("../../../state_of_the_union.txt") as f:state_of_the_union = f.read()

from langchain.text_splitter import SpacyTextSplittertext_splitter = SpacyTextSplitter(chunk_size=1000)texts = text_splitter.split_text(state_of_the_union)

---

texts[0]

Madam Speaker, Madam Vice President, our First Lady and Second Gentleman.Members of Congress and the Cabinet.
...From President Zelenskyy to every Ukrainian, their fearlessness, their courage, their determination, inspires the world.

---

3、SentenceTransformers

SentenceTransformersTokenTextSplitter 是一个专门用于 sentence-transformer 模型 的文本拆分器。
默认行为是 将文本拆分为 适合您想要使用的 sentence transformer 模型的标记窗口的块。

from langchain.text_splitter import SentenceTransformersTokenTextSplitter

splitter = SentenceTransformersTokenTextSplitter(chunk_overlap=0)
text = "Lorem "count_start_and_stop_tokens = 2
text_token_count = splitter.count_tokens(text=text) - count_start_and_stop_tokens
print(text_token_count) # 2

token_multiplier = splitter.maximum_tokens_per_chunk // text_token_count + 1# `text_to_split` does not fit in a single chunk
text_to_split = text * token_multiplier # 514print(f"tokens in text to split: {splitter.count_tokens(text=text_to_split)}")

tokens in text to split: 514

text_chunks = splitter.split_text(text=text_to_split)print(text_chunks[1]) # lorem

---

---

4、NLTK

The Natural Language Toolkit, 或更被知道为 NLTK, 是一套用Python编程语言编写的 用于英语符号和统计自然语言处理（NLP）的库和程序。

在使用 “\n\n” 分割的基础上, 我们使用 NLTK 的 NLTK tokenizers 来分割。

1. 文本如何被分割: 使用 NLTK tokenizer.
2. 块大小如何计算：按 characters 数

# pip install nltk

# This is a long document we can split up.
with open("../../../state_of_the_union.txt") as f:state_of_the_union = f.read()

from langchain.text_splitter import NLTKTextSplittertext_splitter = NLTKTextSplitter(chunk_size=1000)texts = text_splitter.split_text(state_of_the_union)
print(texts[0])

Madam Speaker, Madam Vice President, our First Lady and Second Gentleman....Groups of citizens blocking tanks with their bodies.

---

5、Hugging Face tokenizer

Hugging Face 有很多 tokenizers。

我们使用 Hugging Face tokenizer, GPT2TokenizerFast 来计算tokens 中的文本长度。

1. 文本如何分割: by character passed in
2. 块大小如何计算: 通过 Hugging Face tokenizer计算的 tokens 数量。

from transformers import GPT2TokenizerFast
from langchain.text_splitter import CharacterTextSplittertokenizer = GPT2TokenizerFast.from_pretrained("gpt2")# This is a long document we can split up.
with open("../../../state_of_the_union.txt") as f:state_of_the_union = f.read()text_splitter = CharacterTextSplitter.from_huggingface_tokenizer(tokenizer, chunk_size=100, chunk_overlap=0
)
texts = text_splitter.split_text(state_of_the_union)

---

texts[0]

Madam Speaker, Madam Vice President, our First Lady and Second Gentleman. Members of Congress and the Cabinet. Justices of the Supreme Court. My fellow Americans.  ...With a duty to one another to the American people to the Constitution.

---

2024-04-08（一）