DML_ACTIVATION_SOFTMAX1_OPERATOR_DESC structure (directml.h)

Article
05/19/2023

Performs a softmax activation function on InputTensor, placing the result into the corresponding element of OutputTensor.

For 1-D InputTensor:
// Let x[i] to be the current element in the InputTensor, and j be the total number of elements in the InputTensor
f(x[i]) = exp(x[i]) / sum(exp(x[0]), ..., exp(x[j-1]))

Where exp(x) is the natural exponentiation function.

Important

This API is available as part of the DirectML standalone redistributable package (see Microsoft.AI.DirectML version 1.9 and later. Also see DirectML version history.

Syntax

struct DML_ACTIVATION_SOFTMAX1_OPERATOR_DESC
{
    const DML_TENSOR_DESC* InputTensor;
    const DML_TENSOR_DESC* OutputTensor;
    UINT AxisCount;
    _Field_size_(AxisCount) const UINT* Axes;
};

Members

InputTensor

Type: const DML_TENSOR_DESC*

The input tensor to read from.

OutputTensor

Type: const DML_TENSOR_DESC*

The output tensor to write the results to.

AxisCount

Type: UINT

The number of axes to calculate reduce sum. This field determines the size of the Axes array.

Axes

Type: _Field_size_(AxisCount) const UINT*

The axes along which to reduce the sum. Values must be in the range [0, InputTensor.DimensionCount - 1].

Remarks

This operator is equivalent to DML_ACTIVATION_SOFTMAX_OPERATOR_DESC when AxisCount == 1, and Axes == {DimensionCount - 1}.

Availability

This operator was introduced in DML_FEATURE_LEVEL_5_1.

Tensor constraints

InputTensor and OutputTensor must have the same DataType, DimensionCount, and Sizes.

Tensor support

Tensor	Kind	Supported dimension counts	Supported data types
InputTensor	Input	1 to 8	FLOAT32, FLOAT16
OutputTensor	Output	1 to 8	FLOAT32, FLOAT16

Requirements


Header	directml.h

Share via