What is a synthetic data recipe?
What is it?
A synthetic data recipe explains how to generate a dataset with a model: the prompt that produces the examples, the model used, the target format, the volume and the filters that discard bad examples.
When to use it
- Lack real examples to train or evaluate a model.
- Add rare cases to a dataset.
- Share a reproducible generation method.
Page fields
| Generator prompt | The prompt that produces the examples. |
|---|---|
| Generator model | The model that generates the data. |
| Target format | The format of the produced dataset (chat, instruction, preferences…). |
| Target volume | The number of examples to generate. |
| Filters | The checks that discard failed or duplicate examples. |
How to use it
“Use” tab: download the recipe, run the prompt with the stated model, then check the result with the dataset quality tool.