Functions
ReLU Activation
- Parameters
activationType = ACTIVATION_RELU
- Function graph

- Expression

GELU Activation
- Tanh approximate computing
- Parameters
- activationType = ACTIVATION_GELU
- geluMode = TANH_MODE
- Function graph

- Expression

- Constraints
Not supported for
Atlas 200I/500 A2 inference products .
- Parameters
- Non-approximate computing
- Parameters
- activationType = ACTIVATION_GELU
- geluMode = NONE_MODE
- Function graph

- Expression

is the cumulative distribution function of the Gaussian distribution.
- Parameters
FAST_GELU Activation
- Parameters
activationType = ACTIVATION_FAST_GELU
- Function graph

- Expression

SWISH Activation
- Parameters
- activationType = ACTIVATION_SWISH
- scale
- Function graph
Related to the input parameter scale. When scale is 1:

- Expression

LOG Activation
- Parameters
activationType = ACTIVATION_LOG
- Expression

SWIGLU_FORWARD Activation
- Parameters
- activationType = ACTIVATION_SWIGLU_FORWARD
- dim
- Expression
a and b equalize the input tensor along the last dimension.

- Constraints
dim can be used to perform equalization only along the last dimension. -1 must be passed.
For
Atlas inference products , only 32-bit aligned data is supported.Not supported for
Atlas 200I/500 A2 inference products .
SWIGLU_BACKWARD Activation
- Parameters
- activationType = ACTIVATION_SWIGLU_BACKWARD
- dim
- Expression
x1: input tensor 1.
x2: input tensor 2
a and b equalize x2 along the last dimension.


- Constraints
dim can be used to perform equalization only along the last dimension. -1 must be passed.
Supported only by
Atlas A2 training products /Atlas A2 inference products andAtlas A3 inference products /Atlas A3 training products .
SIGMOID Activation
- Parameters
activationType = ACTIVATION_SIGMOID
- Function graph

- Expression

FASTER_GELU_FORWARD Activation
- Parameters
activationType = ACTIVATION_FASTER_GELU_FORWARD
- Expression

- Constraints
Not supported for
Atlas 200I/500 A2 inference products .