Macha is a dataset of GitHub README files from popular open-source repositories, designed to evaluate how well chunking algorithms handle technical documentation with mixed content types.

Dataset Description

• Documents: 445 GitHub README files
• Questions: 1,812 question-answer pairs
• Domain: Technical Documentation
• Source: Top GitHub repositories by stars

Key Challenges

This dataset tests chunking algorithms on:

• Markdown formatting (headers, lists, code blocks)
• Mixed content (prose, code examples, tables)
• Technical terminology and API references
• Installation instructions and usage examples
• Cross-references and links

Dataset Structure

Corpus Config

Questions Config

Usage

from datasets import load_dataset

# Load corpus
corpus = load_dataset("chonkie-ai/macha", "corpus", split="train")

# Load questions
questions = load_dataset("chonkie-ai/macha", "questions", split="train")

Part of MTCB

Macha is part of the Massive Text Chunking Benchmark (MTCB), a comprehensive benchmark for evaluating RAG chunking strategies.

License

CC-BY-4.0