Hyperparameter Search

Overview

A hyperparameter search lets you easily start multiple trainings at once to quickly test different model option configurations and find the optimal set of options for your custom model.

Hyperparameter searches can be initiated from the model creation form by clicking the Add Hyperparameter Search Configuration button. This saves your current configuration as version 1 and lets you edit the searchable options for version 2. The saved version 1 configuration and the in-progress version 2 configuration will both be viewable from the hyperparameter search summary that appears on the right side of the page.

As you modify the searchable training options, click the Add Hyperparameter Search Configuration button each time you wish to save the current version and edit a new configuration version. Update or delete configurations at any time from the hyperparameter search summary.

Once you have created all of the desired configurations, click Start Training Models to create trainings for each configuration. See below for the list of model options that are currently supported as part of a hyperparameter search.

Searchable Model Options

1. Learning Rate
2. Weight Decay