Toons that are not looney

A compact form of JSON

I’m feeding LLMs with vast quantities of data, and expect them to return the same, and from the start I’ve reached the conclusion that JSON is too verbose for tabular data. Each row will have the same keys, i.e. the same tokens, repeated over and over, increasing cost. Where possible, I’ve used CSV, but that’s not verbose enough, especially if data is not plainly tabular but combines tabular and hierarchical data - which is usually the case.

Enter TOON - the “Token-Oriented Object Notation” which looks like this:

context:
  task: Our favorite hikes together
  location: Boulder
  season: spring_2025
friends[3]: ana,luis,sam
hikes[3]{id,name,distanceKm,elevationGain,companion,wasSunny}:
  1,Blue Lake Trail,7.5,320,ana,true
  2,Ridge Overlook,9.2,540,luis,false
  3,Wildflower Loop,5.1,180,sam,true

It’s sort of in-between CSV and YAML.

I’m not sure it will be usable directly for the small LLMs I’m using without fine-tuning (they know about JSON, they don’t know about TOON), but larger LLMs should be able to infer the format from a prompt.

Check it out!