sayou-chunking 0.3.6

Chunking components for the Sayou Data Platform

sayou-chunking

The Intelligent Text Splitter for Sayou Fabric.

sayou-chunking splits large texts into smaller, semantically meaningful units called Chunks. This is a critical step for RAG (Retrieval-Augmented Generation) systems, as it directly impacts retrieval accuracy.

It goes beyond simple character splitting by offering structure-aware, semantic, and hierarchical chunking strategies.

💡 Core Philosophy

"Context is King."

Blindly cutting text at 500 characters breaks sentences and loses meaning. sayou-chunking aims to preserve context by:

1. Structure Awareness: Respects document headers, tables, and code blocks (especially in Markdown).
2. Semantic Coherence: Groups sentences that belong to the same topic using similarity metrics.
3. Hierarchy: Maintains Parent-Child relationships to retrieve small precise chunks while providing large context to the LLM.

📦 Installation

pip install sayou-chunking

⚡ Quick Start

The ChunkingPipeline provides a unified interface for various splitting strategies.

from sayou.chunking.pipeline import ChunkingPipeline

def run_demo():
    # 1. Initialize Pipeline
    pipeline = ChunkingPipeline()
    pipeline.initialize()

    # 2. Prepare Input (e.g., from Refinery)
    text_content = """
    # Section 1: Introduction
    Chunking is the process of breaking text down.
    
    ## Benefits
    - Better Retrieval
    - Context Preservation
    """
    
    request = {
        "content": text_content,
        "metadata": {"source": "doc.md"},
        "config": {"chunk_size": 50}
    }

    # 3. Run with Strategy ('markdown', 'recursive', 'semantic', etc.)
    chunks = pipeline.run(request, strategy="markdown")

    # 4. Result
    for i, chunk in enumerate(chunks):
        print(f"[{i}] Type: {chunk.metadata.get('semantic_type')}")
        print(f"    Content: {chunk.content}")

if __name__ == "__main__":
    run_demo()

🔑 Key Components

Splitter

• RecursiveSplitter: The standard strategy. Splits by paragraph -> line -> sentence -> word to keep related text together.
• MarkdownSplitter: Aware of Markdown syntax. Splits by headers (#) first, protecting tables and code blocks.
• FixedLengthSplitter: Hard split by character count. Useful when strict token limits are required.
• StructureSplitter: Splits based on user-defined regex patterns (e.g., "Article \d+").
• SemanticSplitter: Uses cosine similarity between sentences to find topic breakpoints.
• ParentDocumentSplitter: Creates large "Parent" chunks for context and small "Child" chunks for retrieval, linking them together.

🤝 Contributing

We welcome contributions for New Strategies (e.g., CodeSplitter for Python/JS) or Integrations with other embedding models for Semantic Splitting.

📜 License

Apache 2.0 License © 2025 Sayouzone