Data Synthesis

Generate training data variations by synthesizing from existing datasets.

This article explains how to use Hyperstack AI Studio’s synthesis tools to generate high-quality training data variations. It covers when and why to synthesize data, how to initiate synthesis from existing datasets, and available filtering options.

Prefer the API?

You can also generate synthetic data programmatically. See the Synthetic Data API reference for all synthesis operations, including request formats, parameters, and example responses.

Why Synthesize Logs?

When working with model outputs that cannot be directly used for training other models, data synthesis becomes necessary. This is particularly relevant when:

• You have saved outputs from a model that you want to use for training
• You need to create variations of your existing data while maintaining its characteristics

Synthesize Logs

You can generate high-quality synthetic training data directly in the AI Studio UI.

1. Go to the Logs & Datasets Page

   Navigate to the Logs & Datasets page and open the Datasets tab.

2. Select Logs to Synthesize

   By default, all logs in the selected dataset will be included. To narrow the selection, apply filters:

   • Tags
   • Models

3. Click “Synthesize Logs” button and confirm the action.

4. Review Results

   After synthesis:

   • You’ll receive a success notification.
   • The logs table will let you toggle between Original and Synthetic versions.

Back to top