How to make AI understand TOON format ?

[seepine]

I tried the description on the official website, but most AI still have a poor understanding, and even the low-level models of Olama 8b 14b are almost unusable.
How can I write prompt words to enable AI to understand TOON format?

[johannschopplich]

Thanks for the report!

Smaller local models (8B–14B) usually don't learn a new format from a long prose description. In practice it works much better if you show a tiny TOON example and ask the model to copy that pattern, instead of explaining the whole spec.

For example, you can start with a prompt like this:

You will receive data in TOON format (YAML-like with explicit array headers).

Example:

```toon
users[3]{id,name,role}:
  1,Alice,admin
  2,Bob,user
  3,Charlie,user
```

Now here is the real data:

```toon
{{YOUR_TOON_DATA_HERE}}
```

Task: {{YOUR_QUESTION_OR_TRANSFORMATION}}.
If you return data, return it only as a ```toon code block using the same structure.

Keeping the example small (2–5 rows) and very concrete helps a lot, especially on 8B/14B models.

For more patterns and ready‑made prompt templates, see the Using TOON with LLMs guide in the docs.