ai.djl.mxnet.engine

Class MxModel

java.lang.Object

ai.djl.BaseModel

ai.djl.mxnet.engine.MxModel

All Implemented Interfaces:

ai.djl.Model, java.lang.AutoCloseable

public class MxModel
extends ai.djl.BaseModel

MxModel is the MXNet implementation of Model.

MxModel contains all the methods in Model to load and process a model. In addition, it provides MXNet Specific functionality, such as getSymbol to obtain the Symbolic graph and getParameters to obtain the parameter NDArrays

Field Summary

Fields inherited from class ai.djl.BaseModel

artifacts, block, dataType, inputData, manager, modelDir, modelName, properties

Method Summary

Modifier and Type	Method and Description
void	cast(ai.djl.ndarray.types.DataType dataType)
void	close()
java.lang.String[]	getArtifactNames()
void	load(java.nio.file.Path modelPath, java.lang.String prefix, java.util.Map<java.lang.String,java.lang.Object> options) Loads the MXNet model from a specified location.
<I,O> ai.djl.inference.Predictor<I,O>	newPredictor(ai.djl.translate.Translator<I,O> translator)
ai.djl.training.Trainer	newTrainer(ai.djl.training.TrainingConfig trainingConfig)
java.lang.String	toString()

Methods inherited from class ai.djl.BaseModel

describeInput, describeOutput, finalize, getArtifact, getArtifact, getArtifactAsStream, getBlock, getDataType, getName, getNDManager, getProperty, paramPathResolver, readParameters, save, setBlock, setDataType, setModelDir, setProperty

Methods inherited from class java.lang.Object

clone, equals, getClass, hashCode, notify, notifyAll, wait, wait, wait

Methods inherited from interface ai.djl.Model

load, load, newInstance, newInstance, newInstance, newInstance, quantize

Method Detail

load

public void load(java.nio.file.Path modelPath,
                 java.lang.String prefix,
                 java.util.Map<java.lang.String,java.lang.Object> options)
          throws java.io.IOException,
                 ai.djl.MalformedModelException

Loads the MXNet model from a specified location.

MXNet engine looks for {MODEL_NAME}-symbol.json and {MODEL_NAME}-{EPOCH}.params files in the specified directory. By default, MXNet engine will pick up the latest epoch of the parameter file. However, users can explicitly specify an epoch to be loaded:

 Map<String, String> options = new HashMap<>()
 options.put("epoch", "3");
 model.load(modelPath, "squeezenet", options);
 

Parameters:

modelPath - the directory of the model

prefix - the model file name or path prefix

options - load model options, see documentation for the specific engine

Throws:

java.io.IOException - Exception for file loading

ai.djl.MalformedModelException

newTrainer

public ai.djl.training.Trainer newTrainer(ai.djl.training.TrainingConfig trainingConfig)

newPredictor

public <I,O> ai.djl.inference.Predictor<I,O> newPredictor(ai.djl.translate.Translator<I,O> translator)

cast

public void cast(ai.djl.ndarray.types.DataType dataType)

getArtifactNames

public java.lang.String[] getArtifactNames()

close

public void close()

toString

public java.lang.String toString()

Overrides:

toString in class java.lang.Object