Model

import com.mindspore.model;

Model defines model in MindSpore for compiling and running.

Public Member Functions

function	Supported At Cloud-side Inference	Supported At Device-side Inference
boolean build(MappedByteBuffer buffer, int modelType, MSContext context, char[] dec_key, String dec_mode)	✕	√
boolean build(Graph graph, MSContext context, TrainCfg cfg)	✕	√
boolean build(MappedByteBuffer buffer, MSContext context)	√	√
boolean build(String modelPath, MSContext context, char[] dec_key, String dec_mode)	✕	√
boolean build(String modelPath, MSContext context)	√	√
boolean predict()	√	√
boolean runStep()	✕	√
boolean resize(List<MSTensor> inputs, int[][] dims)	√	√
List<MSTensor> getInputs()	√	√
List<MSTensor> getOutputs()	√	√
MSTensor getInputsByTensorName(String tensorName)	√	√
MSTensor getOutputByTensorName(String tensorName)	√	√
List<MSTensor> getOutputsByNodeName(String nodeName)	✕	√
List<String> getOutputTensorNames()	√	√
boolean export(String fileName, int quantizationType, boolean isOnlyExportInfer,List<String> outputTensorNames)	✕	√
boolean exportWeightsCollaborateWithMicro(String weightFile, boolean isInference,boolean enableFp16, List changeableWeightNames)	✕	√
List<MSTensor> getFeatureMaps()	✕	√
boolean updateFeatureMaps(List<MSTensor> features)	✕	√
boolean setTrainMode(boolean isTrain)	✕	√
boolean getTrainMode()	✕	√
boolean setLearningRate(float learning_rate)	✕	√
boolean setupVirtualBatch(int virtualBatchMultiplier, float learningRate, float momentum)	✕	√
void free()	√	√
ModelType	√	√

build

public boolean build(Graph graph, MSContext context, TrainCfg cfg)

Compile MindSpore model by computational graph.

• Parameters

  • graph: computational graph.

  • context: compile context.

  • cfg: train config.

• Returns

  Whether the build is successful.

public boolean build(MappedByteBuffer buffer, int modelType, MSContext context, char[] dec_key, String dec_mode)

Compile MindSpore model by computational graph buffer.

• Parameters

  • buffer: computational graph buffer.

  • modelType: computational graph type. Optionally, there are MT_MINDIR_LITE, MT_MINDIR, corresponding to the ms model (exported by the converter_lite tool) and the mindir model (exported by MindSpore or exported by the converter_lite tool), respectively. Only MT_MINDIR_LITE is valid for Device-side Inference and the parameter value is ignored. Cloud-side inference supports ms and mindir model inference, which requires setting the parameter to the option value corresponding to the model. Cloud-side inference support for ms model will be removed in future iterations, and cloud-side inference via mindir model is recommended.

  • context: compile context.

  • dec_key: define the key used to decrypt the ciphertext model. The key length is 16, 24, or 32.

  • dec_mode: define the decryption mode. Options: AES-GCM, AES-CBC.

• Returns

  Whether the build is successful.

public boolean build(final MappedByteBuffer buffer, int modelType, MSContext context)

Compile MindSpore model by computational graph buffer, the default is MindIR model type.

• Parameters

  • buffer: computational graph buffer.

  • modelType: computational graph type. Optionally, there are MT_MINDIR_LITE, MT_MINDIR, corresponding to the ms model (exported by the converter_lite tool) and the mindir model (exported by MindSpore or exported by the converter_lite tool), respectively. Only MT_MINDIR_LITE is valid for Device-side Inference and the parameter value is ignored. Cloud-side inference supports ms and mindir model inference, which requires setting the parameter to the option value corresponding to the model. Cloud-side inference support for ms model will be removed in future iterations, and cloud-side inference via mindir model is recommended.

  • context: compile context.

• Returns

  Whether the build is successful.

public boolean build(String modelPath, int modelType, MSContext context, char[] dec_key, String dec_mode)

Compile MindSpore model by computational graph file.

• Parameters

  • modelPath: computational graph file.

  • modelType: computational graph type. Optionally, there are MT_MINDIR_LITE, MT_MINDIR, corresponding to the ms model (exported by the converter_lite tool) and the mindir model (exported by MindSpore or exported by the converter_lite tool), respectively. Only MT_MINDIR_LITE is valid for Device-side Inference and the parameter value is ignored. Cloud-side inference supports ms and mindir model inference, which requires setting the parameter to the option value corresponding to the model. Cloud-side inference support for ms model will be removed in future iterations, and cloud-side inference via mindir model is recommended.

  • context: compile context.

  • dec_key: define the key used to decrypt the ciphertext model. The key length is 16, 24, or 32.

  • dec_mode: define the decryption mode. Options: AES-GCM, AES-CBC.

• Returns

  Whether the build is successful.

public boolean build(String modelPath, int modelType, MSContext context)

Compile MindSpore model by computational graph file,no decrypt.

• Parameters

  • modelPath: computational graph file.

  • modelType: computational graph type. Optionally, there are MT_MINDIR_LITE, MT_MINDIR, corresponding to the ms model (exported by the converter_lite tool) and the mindir model (exported by MindSpore or exported by the converter_lite tool), respectively. Only MT_MINDIR_LITE is valid for Device-side Inference and the parameter value is ignored. Cloud-side inference supports ms and mindir model inference, which requires setting the parameter to the option value corresponding to the model. Cloud-side inference support for ms model will be removed in future iterations, and cloud-side inference via mindir model is recommended.

  • context: compile context.

• Returns

  Whether the build is successful.

predict

public boolean predict()

Run predict.

• Returns

Whether the predict is successful.

runStep

public boolean runStep()

Run train by step.

• Returns

Whether the run is successful.

resize

public boolean resize(List<MSTensor> inputs, int[][] dims)

Resize inputs shape.

• Parameters

  • inputs: Model inputs.

  • dims: Define the new inputs shape.

• Returns

  Whether the resize is successful.

getInputs

public List<MSTensor> getInputs()

Get the input MSTensors of MindSpore model.

• Returns

  The MindSpore MSTensor list.

getOutputs

public List<MSTensor> getOutputs()

Get the output MSTensors of MindSpore model.

• Returns

  The MindSpore MSTensor list.

getInputsByTensorName

public MSTensor getInputsByTensorName(String tensorName)

Get the input MSTensors of MindSpore model by the node name.

• Parameters

  • tensorName: Define the tensor name.

• Returns

  MindSpore MSTensor.

getOutputsByNodeName

public List<MSTensor> getOutputsByNodeName(String nodeName)

Get the output MSTensors of MindSpore model by the node name.

• Parameters

  • nodeName: Define the node name.

• Returns

  The MindSpore MSTensor list.

getOutputTensorNames

public List<String> getOutputTensorNames()

Get output tensors names of the model compiled by this session.

• Returns

  The vector of string as output tensor names in order.

getOutputByTensorName

public MSTensor getOutputByTensorName(String tensorName)

Get the MSTensors output of MindSpore model by the tensor name.

• Parameters

  • tensorName: Define the tensor name.

• Returns

  MindSpore MSTensor.

export

public boolean export(String fileName, int quantizationType, boolean isOnlyExportInfer,List<String> outputTensorNames)

Export the model.

• Parameters

  • fileName: Model file name.

  • quantization_type: The quant type.

  • isOnlyExportInfer: Is only export infer.

  • outputTensorNames: The output tensor names for export.

• Returns

  Whether the export is successful.

exportweightscollaboratewithmicro

public boolean exportWeightsCollaborateWithMicro(String weightFile, boolean isInference,boolean enableFp16, List<String> changeableWeightNames)

Export the training model weights.

• Parameters

  • weightFile: Model weight file name.

  • isInference: Whether to export the weights of inference graph.

  • enableFp16: Whether to save as float16 dtype.

  • changeableWeightNames: The weight tensor names whose shape is changeable.

• Returns

  Whether the export is successful.

getFeatureMaps

public List<MSTensor> getFeatureMaps()

Get the FeatureMap.

• Returns

  FeatureMaps tensor list.

updatefeaturemaps

public boolean updateFeatureMaps(List<MSTensor> features)

Update model Features.

• Parameters

  • features: New featureMaps tensor List.

• Returns

  Whether the model features is successfully update.

settrainMode

public boolean setTrainMode(boolean isTrain)

Set train mode.

• Parameters

  • isTrain: Is train mode.

gettrainmode

public boolean getTrainMode()

Get train mode.

• Returns

  Whether the model work in train mode.

setLearningRate

public boolean setLearningRate(float learning_rate)

set learning rate.

• Parameters

  • learning_rate: learning rate.

• Returns

  Whether the set learning rate is successful.

setupVirtualBatch

public boolean setupVirtualBatch(int virtualBatchMultiplier, float learningRate, float momentum)

Set the virtual batch.

• Parameters

  • virtualBatchMultiplier: virtual batch multuplier.

  • learningRate: learning rate.

  • momentum: monentum.

• Returns

  Whether the virtual batch is successfully set.

free

public void free()

Free Model.

ModelType

import com.mindspore.config.ModelType;

Model file type.

public static final int MT_MINDIR = 0;
public static final int MT_AIR = 1;
public static final int MT_OM = 2;
public static final int MT_ONNX = 3;
public static final int MT_MINDIR_LITE = 4;