Training | Umnai | Documentation

The platform trains models through induction. Induction learns an explainable model from the features and target in a prepared dataset. Training settings and optional constraints control the learning process and the structure it can produce.

How training works

Dataset and workspace

Choose the dataset created by onboarding and the workspace that will contain the training run and its model.

Learning settings

A training configuration defines the learning behaviour. Optional model constraints guide the modules considered during induction.

Compute and job

A training job combines these choices with a training compute configuration and a runtime limit.

Trained model

When the job completes, inspect the output model and confirm that it is available for deployment.

Training resources and their relationships — A training job uses a dataset, learning settings and training compute to create a model in a workspace.

Learning settings and compute

Use the training configuration as a starting point, then adjust it when you need control over model complexity, feature interactions or learning behaviour. Constraints add explicit guidance about the modules used during training.

Training compute is separate from the processing compute used for ingestion and onboarding. In Model Lab > Training, the configuration, compute and job tabs keep these settings separate from the runs that use them.

A completed job produces a trained model; it does not create a running deployment. Use Serving when the model is ready for evaluation or explanation queries.

Next steps

Prepare training compute and a training configuration. Review model constraints if needed, then create a training job.