1. Learn
    2. /
  2. Courses
    3. /
  3. Developing LLM Applications with LangChain
Connected

Exercise

Recursively splitting by character

Many developers are using a recursive character splitter to split documents based on a specific list of characters. These characters are paragraphs, newlines, spaces, and empty strings, by default: ["\n\n", "\n", " ", ""].

Effectively, the splitter tries to split by paragraphs, checks to see if the chunk_size and chunk_overlap values are met, and if not, splits by sentences, then words, and individual characters.

Often, you'll need to experiment with different chunk_size and chunk_overlap values to find the ones that work well for your documents.

Instructions

100 XP
  • Import the RecursiveCharacterTextSplitter class from langchain_text_splitters.
  • Create a RecursiveCharacterTextSplitter instance with separators=["\n", " ", ""], chunk_size=24, and chunk_overlap=10.
  • Use the .split_text() method to split the quote and print the chunks and chunk lengths.