Table of contentsClick link to navigate to the desired location
This content has been automatically translated from Ukrainian.
TOON format is a simple text data format that simplifies work with JSON-like structures. It is flat: each key and value is written on a separate line, without parentheses, quotation marks and complex structures.
For example, JSON
{
"name": "Yuki",
"age": 28,
"skills": ["ruby", "rails", "ai"]
}
in TOON it looks like
name: Yuki age: 28 skills: ruby, rails, ai
Why does TOON make sense when working with AI?
- Flat structure reduces model errors. JSON is complex: commas, parentheses, quotation marks - models sometimes skip or break them. TOON removes this "noise" and leaves only data.
- Fewer tokens. In large models (GPT, LLaMA, etc.), payment and speed depend on the number of tokens. JSON has a lot of syntax: parentheses, quotation marks, commas, spaces. TOON minimizes this by leaving only keys and values, which saves tokens and allows you to work with larger amounts of data without exceeding the limit.
- It is more convenient to whip and edit. People and models find it easier to read a flat list of keys and values. The model less often generates syntax errors and does not "break" the structure.
- Simple storage and analysis. TOON is easier to compare in versions, make diffs, integrate into pipelines where JSON is too bulky.
In short. TOON reduces the tokenized weight of data, makes it easier for models, and at the same time retains the ability to easily convert back into structured data. This is the optimal format for AI-configs, prompts and large arrays of structured data.
This post doesn't have any additions from the author yet.
