LangChain

Langchain

An open source framework for building GenAI/LLM applications

• Python and Javascript packages
• Focused on composition and modularity
• Key value adds
  • modular components
  • Use cases: common ways to combine components
• diagram

pipenv install langchain
pipenv install langchain-openai
pipenv install langchain-community
pipenv install langchainhub

• langchain-openai: The 3rd party package which integrates langchain with OpenAI’s LLMs
• langchain-community: The package that holds a lot of community contributed code like text splitters and output parsers
• langchainhub: Extension of LangChain, helps us to download prompts dynamically contributed by the community
• They are separate so we don’t have to download bloat

runnables

• Each runnable is a self-contained unit that performs a specific task
• You can chain these units together (| operator) to define a workflow
  • RunnableParallel
    • executes multiple Runnable operations in parallel and aggregates their outputs into a single dictionary
    • used when needed to perform independent tasks concurrently
  • RunnablePassthrough
    • simply passes the input it receives unchanged to the next step in the chain

Components

• Models, prompts, parsers
  • Models
    • LLMs: 20+ integrations
    • Chat models
    • Text embedding models: 10+ integrations
  • Prompts
    • PromptTemplate
    • ChatPromptTemplate
    • Output parsers: 5+ implementations
      • Retry/fixing logic
    • Example selectors: 5+ implementations
• LCEL (LangChain Expression Language)
  • Chains in langchain (outdated now)
• With OpenAI
  • OpenAI Function Calling - general idea
  • 📌OpenAI Function Calling with LCEL (+pydantic)
    • Pydantic
• 📌Tagging and Extraction
• 📌Tools and Routing
• 📌Conversational Agent

Text Splitter

Other Useful stuff

• load_summarize_chain