Carregando e dividindo arquivos de código

Retrieval Augmented Generation (RAG) com LangChain

Meri Nova

Machine Learning Engineer

Mais carregadores de documentos...

Uma seleção de diferentes formatos de arquivo.

O markdown bruto usado para criar o arquivo README no repositório do LangChain no GitHub.

Markdown renderizado do arquivo README.md do repositório do LangChain no GitHub.

Carregando arquivos Markdown (.md)

from langchain_community.document_loaders import UnstructuredMarkdownLoader


loader = UnstructuredMarkdownLoader("README.md")

markdown_content = loader.load()
print(markdown_content[0])

Document(page_content='# Discord Text Classification ![Python Version](https...'
         metadata={'source': 'README.md'})

Carregando arquivos Python (.py)

from abc import ABC, abstractmethod

class LLM(ABC):
  @abstractmethod
  def complete_sentence(self, prompt):
    pass

...

Integrado em apps de RAG para escrever/corrigir código, criar docs etc.
Imports, classes, funções etc.

from langchain_community.document_loaders \
    import PythonLoader

loader = PythonLoader('chatbot.py')


python_data = loader.load()
print(python_data[0])

Document(page_content='from abc import ABC, ...

class LLM(ABC):
  @abstractmethod
...',
metadata={'source': 'chatbot.py'})

Dividindo arquivos de código

python_splitter = RecursiveCharacterTextSplitter(
    chunk_size=150, chunk_overlap=10
)


chunks = python_splitter.split_documents(python_data)
for i, chunk in enumerate(chunks[:3]):
    print(f"Chunk {i+1}:\n{chunk.page_content}\n")

Chunk 1:
from abc import ABC, abstractmethod

class LLM(ABC):
  @abstractmethod
  def complete_sentence(self, prompt):
    pass

Chunk 2:
class OpenAI(LLM):
  def complete_sentence(self, prompt):
    return prompt + " ... OpenAI end of sentence."

class Anthropic(LLM):

Chunk 3:
def complete_sentence(self, prompt):
    return prompt + " ... Anthropic end of sentence."

Dividindo por linguagem

separators
- ["\n\n", "\n", " ", ""]
- ["\nclass ", "\ndef ", "\n\tdef ", "\n\n", " ", ""]

from langchain_text_splitters import RecursiveCharacterTextSplitter, Language

python_splitter = RecursiveCharacterTextSplitter.from_language(

    language=Language.PYTHON, chunk_size=150, chunk_overlap=10

)


chunks = python_splitter.split_documents(data)


for i, chunk in enumerate(chunks[:3]):
    print(f"Chunk {i+1}:\n{chunk.page_content}\n")

Chunk 1:
from abc import ABC, abstractmethod

Chunk 2:
class LLM(ABC):
  @abstractmethod
  def complete_sentence(self, prompt):
    pass

Chunk 3:
class OpenAI(LLM):
  def complete_sentence(self, prompt):

Vamos praticar!

Retrieval Augmented Generation (RAG) com LangChain