ActivationType

Description

Enumeration of activation function types.

Definition

enum ActivationType : int {
    ACTIVATION_UNDEFINED = 0,
    ACTIVATION_RELU,
    ACTIVATION_GELU,
    ACTIVATION_FAST_GELU,
    ACTIVATION_SWISH,
    ACTIVATION_LOG,
    ACTIVATION_SWIGLU_FORWARD,
    ACTIVATION_SWIGLU_BACKWARD,
    ACTIVATION_SIGMOID,
    ACTIVATION_FASTER_GELU_FORWARD,
    ACTIVATION_MAX,
};

Members

Member

Description

ACTIVATION_UNDEFINED

Undefined type

ACTIVATION_RELU

RELU activation type

ACTIVATION_GELU

GELU activation type

ACTIVATION_FAST_GELU

FAST_GELU activation type

Fast version of the GELU activation function. The approximate GELU activation function is executed for each element in a tensor to speed the calculation with higher accuracy.

ACTIVATION_SWISH

SWISH activation type

ACTIVATION_LOG

LOG activation type

ACTIVATION_SWIGLU_FORWARD

SWIGLU_FORWARD activation type.

SwiGLU forward activation function. For the Atlas inference products, only 32-bit aligned data is supported.

ACTIVATION_SWIGLU_BACKWARD

SWIGLU_BACKWARD activation type.

Backward of the SwigLU forward activation function, which is used for gradient calculation. It is supported only by the Atlas A2 inference products.

ACTIVATION_SIGMOID

SIGMOID activation type

ACTIVATION_FASTER_GELU_FORWARD

FASTER_GELU_FORWARD activation type

ACTIVATION_MAX

Maximum enumerated value, inactive type