Module activation

The activation module.

Functions

gelu

Applies the Gaussian Error Linear Units function as described in the paper Gaussian Error Linear Units (GELUs).

hard_sigmoid

Applies the hard sigmoid function

leaky_relu

Applies the leaky rectified linear unit function.

log_sigmoid

Applies the log sigmoid function.

log_softmax

Applies the log softmax function on the input tensor along the given dimension.

mish

Applies the Mish function as described in the paper in Mish: A Self Regularized Non-Monotonic Neural Activation Function.

prelu

Applies Parametric ReLu activation function as described in the paper Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification. PReLu(x) = max(0,x) + \alpha * min(0,x) tensor is assumed to be of shape [batch_size, channels, …] alpha is assumed to be of shape [channels] or [1]

quiet_softmax

Applies the “quiet softmax” function on the input tensor along the given dimension. This function is similar to the softmax function, but it allows for “no selection”, e.g., all outputs can tend to zero.

relu

Applies the rectified linear unit function as described in the paper Deep Learning using Rectified Linear Units (ReLU).

sigmoid

Applies the sigmoid function.

silu

Applies the silu function

softmax

Applies the softmax function on the input tensor along the given dimension.

softmin

Applies the softmin function on the input tensor along the given dimension.

softplus

Applies the softplus function

tanh

Applies the tanh function