torch (version 0.14.2)

nn_gelu: GELU module

Description

Applies the Gaussian Error Linear Units function: $$\mbox{GELU}(x) = x * \Phi(x)$$

Usage

nn_gelu(approximate = "none")

Arguments

approximate

the gelu approximation algorithm to use: 'none' or 'tanh'. Default: 'none'.

Shape

• Input: \((N, *)\) where * means, any number of additional dimensions

• Output: \((N, *)\), same shape as the input

Details

where \(\Phi(x)\) is the Cumulative Distribution Function for Gaussian Distribution.

Examples

if (torch_is_installed()) {
m <- nn_gelu()
input <- torch_randn(2)
output <- m(input)
}