Model training integration

This page contains a step-by-step guide for training a model with your labeled data with one click in Labelbox.

After you create a model and create a model run specifying data rows and data splits, you'll notice that the Train model button is still inactivated and grey. This is because you need to establish model training settings before you can start with model training.

Before you start

Before you can use the one-click model training integration, make sure you set up your model training integration first. You only need to connect your training cloud account to your Labelbox organization once. Then, you can use the same integration for all model training requests.

📘

Example datasets

To ensure a smooth experience with this tutorial, we recommend using one of our example datasets.
You can use any of the dataset seeding scripts in this Github repo.

Below are the annotation types that are supported directly by our reference models.

Step 1: Configure the model training settings

1. To configure the settings for your model training integration, go to the Settings tab and click Model training.

2. In the URL, replace <your_ip> with the IP address of your training pipeline coordinator deployed in the cloud (it should be printed in the console once at the end of the deploy script).

3. Replace Secret with the SERVICE_SECRET you set in your .env file.

4. Then, click Save.

Step 2: Select the machine learning task

1. Now that you have the integration settings configured, go back to the Model runs tab.

2. The Train model button should now be clickable. Click Train model to select one of the available models.

3. You will see a list of reference models provided by Labelbox. If you set up your custom training pipeline, those should appear here as well.

4. Once you select the model, click Submit to start model training.

Step 3: View results

The model training will take a few hours to complete. You can come back to the Model runs page to check its progress.

Once the training is complete, you will see the model's predictions on each data row and quantitative evaluation results on this page.

From here, you can evaluate your model , find model errors to improve your next model iteration, find label errors and reduce your labeling budget.