InferenceInput.Builder

Builder

public Builder (InferenceInput.Params params, 
                byte[] inputData)

Creates a new Builder with provided runtime parameters and input data.

For LiteRT, inputData field is mapped to inputs of runForMultipleInputsOutputs: https://www.tensorflow.org/lite/api_docs/java/org/tensorflow/lite/InterpreterApi#parameters_9 For example, if a model takes multiple inputs:

String[] input0 = {"foo", "bar"}; // string tensor shape is [2].
 int[] input1 = new int[]{3, 2, 1}; // int tensor shape is [3].
 Object[] data = {input0, input1, ...};
 byte[] inputData = serializeObject(data);
 

For Executorch, inputData field is mapped to a serialized EValue array.

Builder

public Builder (InferenceInput.Params params, 
                Object[] inputData, 
                InferenceOutput expectedOutputStructure)

Note: use InferenceInput.Builder.Builder(Params, byte[]) instead.

Creates a new Builder for LiteRT model inference input. For LiteRT, inputData field is mapped to inputs of runForMultipleInputsOutputs: https://www.tensorflow.org/lite/api_docs/java/org/tensorflow/lite/InterpreterApi#parameters_9 The inputs should be in the same order as inputs * of the model. *

For example, if a model takes multiple inputs: *

String[] input0 = {"foo", "bar"}; // string tensor shape is [2].
 int[] input1 = new int[]{3, 2, 1}; // int tensor shape is [3].
 Object[] inputData = {input0, input1, ...};
 

If a model produce string tensors:

String[] output = new String[3][2];  // Output tensor shape is [3, 2].
 HashMap<Integer, Object> outputs = new HashMap<>();
 outputs.put(0, output);
 expectedOutputStructure = new InferenceOutput.Builder().setDataOutputs(outputs).build();

 

build

public InferenceInput build ()

Builds the instance. This builder should not be touched after calling this!

setBatchSize

public InferenceInput.Builder setBatchSize (int value)

The number of input examples. Adopter can set this field to run batching inference. The batch size is 1 by default. The batch size should match the input data size.

setExpectedOutputStructure

public InferenceInput.Builder setExpectedOutputStructure (InferenceOutput value)

The empty InferenceOutput representing the expected output structure. It's only required by LiteRT model. For LiteRT, the inference code will verify whether this expected output structure matches model output signature.

If a model produce string tensors:

String[] output = new String[3][2];  // Output tensor shape is [3, 2].
 HashMap<Integer, Object> outputs = new HashMap<>();
 outputs.put(0, output);
 expectedOutputStructure = new InferenceOutput.Builder().setDataOutputs(outputs).build();
 

setInputData

public InferenceInput.Builder setInputData (byte[] value)

A byte array that holds input data. The inputs should be in the same order as inputs of the model.

For LiteRT, this field is mapped to inputs of runForMultipleInputsOutputs: https://www.tensorflow.org/lite/api_docs/java/org/tensorflow/lite/InterpreterApi#parameters_9

String[] input0 = {"foo", "bar"}; // string tensor shape is [2].
 int[] input1 = new int[]{3, 2, 1}; // int tensor shape is [3].
 Object[] data = {input0, input1, ...};
 byte[] inputData = serializeObject(data);
 

For Executorch model, this field is a serialized EValue array.

setInputData

public InferenceInput.Builder setInputData (Object... value)

Note: use InferenceInput.Builder.setInputData(byte[]) instead.

An array of input data. The inputs should be in the same order as inputs of the model.

For example, if a model takes multiple inputs:

String[] input0 = {"foo", "bar"}; // string tensor shape is [2].
 int[] input1 = new int[]{3, 2, 1}; // int tensor shape is [3].
 Object[] inputData = {input0, input1, ...};
 

setParams

public InferenceInput.Builder setParams (InferenceInput.Params value)

The configuration that controls runtime interpreter behavior.