API Reference

KerasModelImport API — all import methods and configuration options

Keras Model Import API Reference

KerasModelImport is the main entry point for importing Keras models into DL4J. It reads stored Keras configurations and weights from:

• A single HDF5 archive storing model architecture, training configuration, and weights (produced by model.save())

• A JSON file containing the model configuration combined with a separate HDF5 file containing weights (produced by model.to_json() and model.save_weights())

Source

---

Importing Functional API (ComputationGraph) Models

These methods import Keras Functional API models (keras.models.Model) as DL4J ComputationGraph.

---

importKerasModelAndWeights (InputStream)

public static ComputationGraph importKerasModelAndWeights(
        InputStream modelHdf5Stream,
        boolean enforceTrainingConfig)
        throws IOException, UnsupportedKerasConfigurationException, InvalidKerasConfigurationException

Load a Keras Functional API model saved using model.save(...) from an InputStream.

Parameters:

• modelHdf5Stream — InputStream containing the HDF5 archive

• enforceTrainingConfig — whether to enforce training configuration options; set to false if the model was not compiled

Returns: ComputationGraph

---

public static ComputationGraph importKerasModelAndWeights(InputStream modelHdf5Stream)
        throws IOException, UnsupportedKerasConfigurationException, InvalidKerasConfigurationException

Load a Keras Functional API model from an InputStream with training configuration enforced by default.

Parameters:

• modelHdf5Stream — InputStream containing the HDF5 archive

Returns: ComputationGraph

---

importKerasModelAndWeights (file path)

public static ComputationGraph importKerasModelAndWeights(
        String modelHdf5Filename,
        int[] inputShape,
        boolean enforceTrainingConfig)
        throws IOException, UnsupportedKerasConfigurationException, InvalidKerasConfigurationException

Load a Keras Functional API model saved using model.save(...) from a file path. Use this overload for models that do not include input shape information (e.g., include_top=False models).

Parameters:

• modelHdf5Filename — path to the HDF5 archive

• inputShape — optional input shape; pass null if the model already contains shape information

• enforceTrainingConfig — whether to enforce training configuration options

Returns: ComputationGraph

---

public static ComputationGraph importKerasModelAndWeights(
        String modelHdf5Filename,
        boolean enforceTrainingConfig)
        throws IOException, UnsupportedKerasConfigurationException, InvalidKerasConfigurationException

Parameters:

• modelHdf5Filename — path to the HDF5 archive

• enforceTrainingConfig — whether to enforce training configuration options

Returns: ComputationGraph

---

public static ComputationGraph importKerasModelAndWeights(String modelHdf5Filename)
        throws IOException, UnsupportedKerasConfigurationException, InvalidKerasConfigurationException

Load a Keras Functional API model from a file path with training configuration enforced by default.

Parameters:

• modelHdf5Filename — path to the HDF5 archive

Returns: ComputationGraph

---

importKerasModelAndWeights (JSON config + weights)

public static ComputationGraph importKerasModelAndWeights(
        String modelJsonFilename,
        String weightsHdf5Filename,
        boolean enforceTrainingConfig)
        throws IOException, InvalidKerasConfigurationException, UnsupportedKerasConfigurationException

Load a Keras Functional API model for which the configuration and weights were saved separately using model.to_json() and model.save_weights(...).

Parameters:

• modelJsonFilename — path to the JSON file storing the model configuration

• weightsHdf5Filename — path to the HDF5 archive storing model weights

• enforceTrainingConfig — whether to enforce training configuration options

Returns: ComputationGraph

---

public static ComputationGraph importKerasModelAndWeights(
        String modelJsonFilename,
        String weightsHdf5Filename)
        throws IOException, InvalidKerasConfigurationException, UnsupportedKerasConfigurationException

Parameters:

• modelJsonFilename — path to the JSON file

• weightsHdf5Filename — path to the HDF5 weights file

Returns: ComputationGraph

---

importKerasModelConfiguration

public static ComputationGraphConfiguration importKerasModelConfiguration(
        String modelJsonFilename,
        boolean enforceTrainingConfig)
        throws IOException, InvalidKerasConfigurationException, UnsupportedKerasConfigurationException

Load only the model configuration (no weights) for a Keras Functional API model.

Parameters:

• modelJsonFilename — path to the JSON file

• enforceTrainingConfig — whether to enforce training configuration options

Returns: ComputationGraphConfiguration

---

public static ComputationGraphConfiguration importKerasModelConfiguration(String modelJsonFilename)
        throws IOException, InvalidKerasConfigurationException, UnsupportedKerasConfigurationException

Parameters:

• modelJsonFilename — path to the JSON file

Returns: ComputationGraphConfiguration

---

Importing Sequential (MultiLayerNetwork) Models

These methods import Keras Sequential models as DL4J MultiLayerNetwork.

---

importKerasSequentialModelAndWeights (InputStream)

public static MultiLayerNetwork importKerasSequentialModelAndWeights(
        InputStream modelHdf5Stream,
        boolean enforceTrainingConfig)
        throws IOException, InvalidKerasConfigurationException, UnsupportedKerasConfigurationException

Load a Keras Sequential model saved using model.save(...) from an InputStream.

Parameters:

• modelHdf5Stream — InputStream containing the HDF5 archive

• enforceTrainingConfig — whether to enforce training configuration options

Returns: MultiLayerNetwork

---

public static MultiLayerNetwork importKerasSequentialModelAndWeights(InputStream modelHdf5Stream)
        throws IOException, InvalidKerasConfigurationException, UnsupportedKerasConfigurationException

Parameters:

• modelHdf5Stream — InputStream containing the HDF5 archive

Returns: MultiLayerNetwork

---

importKerasSequentialModelAndWeights (file path)

public static MultiLayerNetwork importKerasSequentialModelAndWeights(
        String modelHdf5Filename,
        int[] inputShape,
        boolean enforceTrainingConfig)
        throws IOException, InvalidKerasConfigurationException, UnsupportedKerasConfigurationException

Load a Keras Sequential model from a file path. Supply inputShape for models that omit input shape in their configuration.

Parameters:

• modelHdf5Filename — path to the HDF5 archive

• inputShape — optional input shape array; pass null if not needed

• enforceTrainingConfig — whether to enforce training configuration options

Returns: MultiLayerNetwork

---

public static MultiLayerNetwork importKerasSequentialModelAndWeights(
        String modelHdf5Filename,
        boolean enforceTrainingConfig)
        throws IOException, InvalidKerasConfigurationException, UnsupportedKerasConfigurationException

Parameters:

• modelHdf5Filename — path to the HDF5 archive

• enforceTrainingConfig — whether to enforce training configuration options

Returns: MultiLayerNetwork

---

public static MultiLayerNetwork importKerasSequentialModelAndWeights(String modelHdf5Filename)
        throws IOException, InvalidKerasConfigurationException, UnsupportedKerasConfigurationException

Parameters:

• modelHdf5Filename — path to the HDF5 archive

Returns: MultiLayerNetwork

---

importKerasSequentialModelAndWeights (JSON config + weights)

public static MultiLayerNetwork importKerasSequentialModelAndWeights(
        String modelJsonFilename,
        String weightsHdf5Filename,
        boolean enforceTrainingConfig)
        throws IOException, InvalidKerasConfigurationException, UnsupportedKerasConfigurationException

Load a Keras Sequential model for which the configuration and weights were saved separately.

Parameters:

• modelJsonFilename — path to the JSON configuration file

• weightsHdf5Filename — path to the HDF5 weights file

• enforceTrainingConfig — whether to enforce training configuration options

Returns: MultiLayerNetwork

---

public static MultiLayerNetwork importKerasSequentialModelAndWeights(
        String modelJsonFilename,
        String weightsHdf5Filename)
        throws IOException, InvalidKerasConfigurationException, UnsupportedKerasConfigurationException

Parameters:

• modelJsonFilename — path to the JSON configuration file

• weightsHdf5Filename — path to the HDF5 weights file

Returns: MultiLayerNetwork

---

importKerasSequentialConfiguration

public static MultiLayerConfiguration importKerasSequentialConfiguration(
        String modelJsonFilename,
        boolean enforceTrainingConfig)
        throws IOException, InvalidKerasConfigurationException, UnsupportedKerasConfigurationException

Load only the model configuration for a Keras Sequential model. No weights are loaded.

Parameters:

• modelJsonFilename — path to the JSON configuration file

• enforceTrainingConfig — whether to enforce training configuration options

Returns: MultiLayerConfiguration

---

public static MultiLayerConfiguration importKerasSequentialConfiguration(String modelJsonFilename)
        throws IOException, InvalidKerasConfigurationException, UnsupportedKerasConfigurationException

Parameters:

• modelJsonFilename — path to the JSON configuration file

Returns: MultiLayerConfiguration

---

Exception Types

Exception	Meaning
InvalidKerasConfigurationException	The Keras configuration contains elements that are recognized but cannot be parsed or mapped
UnsupportedKerasConfigurationException	The Keras configuration contains features that are not supported by the importer
IOException	File not found or unreadable

---

enforceTrainingConfig Flag

The enforceTrainingConfig parameter controls how training-related configuration in the HDF5 is handled:

• true (default): an exception is thrown if any training configuration element cannot be parsed or mapped. Use this when you intend to continue training the model in DL4J.

• false: training configuration parsing errors are logged as warnings but do not stop import. Use this for inference-only workflows or when the Keras model was not compiled before saving.

---

Method Selection Guide

Situation	Method to use
Sequential model, single .h5 file	importKerasSequentialModelAndWeights(String)
Sequential model, config + weights separate	importKerasSequentialModelAndWeights(String, String)
Sequential model, config only	importKerasSequentialConfiguration(String)
Sequential model, from InputStream	importKerasSequentialModelAndWeights(InputStream)
Functional model, single .h5 file	importKerasModelAndWeights(String)
Functional model, config + weights separate	importKerasModelAndWeights(String, String)
Functional model, config only	importKerasModelConfiguration(String)
Functional model, from InputStream	importKerasModelAndWeights(InputStream)
Model missing input shape	Use the overload with int[] inputShape parameter
Model not compiled in Keras	Pass enforceTrainingConfig=false