.. default-domain:: cpp
The softmax primitive performs softmax along a particular axis on data with arbitrary dimensions. All other axes are treated as independent (batch).
In general form, the operation is defined by the following formulas. The variable names follow the standard :ref:`conventions-label`.
When the specified algorithm is softmax:
\dst(\overline{ou}, c, \overline{in}) = \frac {e^{\src(\overline{ou}, c, \overline{in}) - \nu(\overline{ou}, \overline{in})}} { \sum\limits_{ic} e^{\src(\overline{ou}, ic, \overline{in}) - \nu(\overline{ou}, \overline{in})} }.
When the specified algorithm is logsoftmax, the following numerically stable formula is used:
\dst(\overline{ou}, c, \overline{in}) = \ln\left({\frac { e^{\src(\overline{ou}, c, \overline{in}) - \nu(\overline{ou}, \overline{in})} } { \sum\limits_{ic} e^{\src(\overline{ou}, ic, \overline{in}) - \nu(\overline{ou}, \overline{in})} }}\right) = \left(\src(\overline{ou}, c, \overline{in}) - \nu(\overline{ou}, \overline{in})\right) - \ln\left( \sum\limits_{ic} e^{\src(\overline{ou}, ic, \overline{in}) - \nu(\overline{ou}, \overline{in})} \right)
where
- c axis over which the softmax computation is computed on,
- \overline{ou} is the outermost index (to the left of softmax axis),
- \overline{in} is the innermost index (to the right of softmax axis), and
- \nu is used to produce more accurate results and defined as:
\nu(\overline{ou}, \overline{in}) = \max\limits_{ic} \src(\overline{ou}, ic, \overline{in})
There is no difference between the |forward_training| and |forward_inference| propagation kinds.
The backward propagation computes \diffsrc(ou, c, in), based on \diffdst(ou, c, in) and \dst(ou, c, in).
When executed, the inputs and outputs should be mapped to an execution argument index as specified by the following table.
Primitive input/output | Execution argument index |
---|---|
\src | |DNNL_ARG_SRC| |
\dst | |DNNL_ARG_DST| |
\diffsrc | |DNNL_ARG_DIFF_SRC| |
\diffdst | |DNNL_ARG_DIFF_DST| |
- Both forward and backward propagation support in-place operations, meaning
that
src
can be used as input and output for forward propagation, anddiff_dst
can be used as input and output for backward propagation. In case of in-place operation, the original data will be overwritten.
The softmax primitive does not have to support any post-ops or attributes.
The softmax primitive supports the following combinations of data types.
Note
Here we abbreviate data types names for readability. For example, |_f32| is abbreviated to |f32|.
Propagation | Source / Destination |
---|---|
forward / backward | |bf16|, |f32| |
forward | |f16| |
The softmax primitive works with arbitrary data tensors. There is no special meaning associated with any logical dimensions. However, the softmax axis is typically referred to as channels (hence in formulas we use c).
.. doxygenstruct:: dnnl::softmax_forward :project: oneDNN :members:
.. doxygenstruct:: dnnl::softmax_backward :project: oneDNN :members: