Package ai.onnxruntime

Class TensorInfo

java.lang.Object
ai.onnxruntime.TensorInfo
All Implemented Interfaces:
ValueInfo
public class TensorInfo extends Object implements ValueInfo
Describes an OnnxTensor, including it's size, shape and element type.

  • Field Details

    • MAX_DIMENSIONS

      public static final int MAX_DIMENSIONS
      Maximum number of dimensions supported by the Java interface methods.
      See Also:

    • type

      public final OnnxJavaType type
      The Java type of this tensor.

    • onnxType

      public final TensorInfo.OnnxTensorType onnxType
      The native type of this tensor.

  • Method Details

    • getShape

      public long[] getShape()
      Get a copy of the tensor's shape.
      Returns:
      A copy of the tensor's shape.

    • getDimensionNames

      public String[] getDimensionNames()
      Get a copy of the tensor's named dimensions.
      Returns:
      A copof the tensor's named dimensions.

    • toString

      public String toString()
      Overrides:
      toString in class Object

    • isScalar

      public boolean isScalar()
      Returns true if the shape represents a scalar value (i.e. it has zero dimensions).
      Returns:
      True if the shape is a scalar.

    • getNumElements

      public long getNumElements()
      Returns the number of elements in this tensor.

      If the returned value is negative, then this tensor info refers to an input or output placeholder which has symbolic dimensions, and the element count cannot be computed without specifying the symbolic dimensions.

      Returns:
      The number of elements.

    • makeCarrier

      public Object makeCarrier() throws OrtException
      Constructs an array the right shape and type to hold this tensor.

      Note for String tensors, this carrier is a single dimensional array with enough space for all elements as that's the expected format of the native code. It can be reshaped to the correct shape using OrtUtil.reshape(String[],long[]).

      For fp16 and bf16 tensors the output carrier type is float, and so this method produces multidimensional float arrays.

      Returns:
      A multidimensional array of the appropriate primitive type (or String).
      Throws:
      OrtException - If the shape isn't representable in Java (i.e. if one of its indices is greater than an int).

    • constructFromJavaArray

      public static TensorInfo constructFromJavaArray(Object obj) throws OrtException
      Constructs a TensorInfo from the supplied multidimensional Java array, used to allocate the appropriate amount of native memory.
      Parameters:
      obj - The object to inspect.
      Returns:
      A TensorInfo which can be used to make the right size Tensor.
      Throws:
      OrtException - If the supplied Object isn't an array, or is an invalid type.

    • constructFromBuffer

      public static TensorInfo constructFromBuffer(Buffer buffer, long[] shape, OnnxJavaType type) throws OrtException
      Constructs a TensorInfo from the supplied byte buffer.
      Parameters:
      buffer - The buffer to inspect.
      shape - The shape of the tensor.
      type - The Java type.
      Returns:
      A TensorInfo for a tensor.
      Throws:
      OrtException - If the supplied buffer doesn't match the shape.

    • constructFromSparseTensor

      public static <T extends Buffer> TensorInfo constructFromSparseTensor(OnnxSparseTensor.SparseTensor<T> tensor) throws OrtException
      Constructs a TensorInfo from the supplied OnnxSparseTensor.SparseTensor.
      Type Parameters:
      T - The buffer type.
      Parameters:
      tensor - The sparse tensor.
      Returns:
      A TensorInfo for a sparse tensor.
      Throws:
      OrtException - If the supplied tensor has too many elements for it's shape.