Zero-Overhead Notation format

From Wiki, the free encyclopedia
Lightweight data serialization format optimized for LLM token usage

Zero-Overhead Notation (ZON) is a lightweight data serialization format created by Roni Bhakta. It was designed explicitly to minimize token usage when interacting with Large Language Models (LLMs) such as those developed by OpenAI and Google.

Overview

[edit]

When passing strictly formatted data (like JSON or YAML) into the context window of highly complex LLMs, the models consume a massive amount of "tokens" just parsing syntax limiters (brackets, quotation marks, commas). ZON significantly minimizes these limiters, relying on a syntax layout that is inherently easier for an LLM to tokenize without losing structural integrity.

In production benchmarks, ZON has been shown to reduce token consumption by approximately 50% compared to standard JSON structures, while boasting a remarkably high retrieval accuracy rate of 98.6%.

Features

[edit]
  • Schema Validation: Supported across Python and TypeScript implementations.
  • Streaming: Allows parsing ZON formatted files on-the-fly, particularly useful when reading chunks of output directly from LLM streaming APIs.
  • Human-Readable: Retains readability despite dropping syntactic limiters common in JSON.

See also

[edit]