Model Definitions

The model definition is the interface between Determined and the user’s application framework (e.g., Keras, TensorFlow), in terms of loading training data, describing a model architecture, and specifying the underlying iterative optimization training algorithms.

There are two interfaces you can use to write model definitions: Native and Trial. With the Native API (TensorFlow Keras and Estimators), developers instrument training scripts adhering to these frameworks’ native patterns including their training loops. With the Trial API, developers expose the components of a model definition to Determined’s training loop implementation.

The Python SDK supports the Native API for TF Keras and Estimators. The command line interface and Python SDK support the Trial API.

To read more about the Native API, see:

• Native API Tutorial

• Native API

To read more about the Trial API, see:

• Trial API

• determined.pytorch.PyTorchTrial

• determined.keras.TFKerasTrial

• determined.estimator.EstimatorTrial

Best Practices

To learn more about some best practices when writing Trial API model definitions, see Best Practices for Model Definitions.