Fine-tune a model

Train a custom model on your own data, then deploy it and test it in the Playground.

See the Model Fine-Tuning Guide for more details.

Before you begin

You'll need:

1. A Hyperstack account. Sign up on the registration page using your email or an SSO provider (Google, Microsoft, or GitHub).
2. Billing set up. Add credit to your account.
3. Access to AI Studio. Open AI Studio in the Hyperstack console.
4. Training data. Example conversations in .jsonl format, with at least 10 valid logs.

Fine-tune, deploy, and test

1. Upload your training data

   On the Data page, click Upload Logs, select your .jsonl file, add at least one tag, and click Validate and upload. For the file format and other options, see Logs. To group multiple uploads for a single training run, create a dataset.

2. Start a fine-tuning job

   On the My Models page, click Fine-Tune a model. Name your model and select a base model. To train on the logs you uploaded, set the training data to By Tags (or All Logs) instead of the default sample dataset. Review the cost and time estimate, then click Create. See Fine-tuning for the full set of options.

3. Monitor and evaluate training

   Open your model from the My Models page to track training progress and review loss metrics. See Training Metrics to interpret the results, and Evaluations to benchmark your model.

4. Deploy your model

   On the model's detail page, use the Deploy toggle to bring the model online. Deployment takes a few seconds. See Deployment for details.

5. Test it in the Playground

   Open the Text Playground, select your deployed model from the model dropdown, and send a prompt. Turn on the Compare side-by-side toggle to test it against the base model.

Next steps

Model Fine-Tuning Guide

A detailed, end-to-end tutorial with a sample dataset.

Evaluate your model

Run benchmarks and custom evaluations on your fine-tuned model.

Model aliases

Give deployed models stable names for your application code.

Overview

Browse the full AI Studio documentation.