ai.djl.nn.norm

Class LayerNorm.Builder

java.lang.Object

ai.djl.nn.norm.LayerNorm.Builder

Enclosing class:

LayerNorm

public static final class LayerNorm.Builder
extends java.lang.Object

The Builder to construct a LayerNorm.

Method Summary

Modifier and Type	Method and Description
LayerNorm.Builder	axis(int... axis) List the axis over which the mean and variance will be calculated (alternative to normalizedShape).
LayerNorm	build() Builds a LayerNorm block.
LayerNorm.Builder	optCenter(boolean val) If True, add offset of `beta` to normalized tensor.
LayerNorm.Builder	optEpsilon(float val) Sets the epsilon value to prevent division by 0.
LayerNorm.Builder	optScale(boolean val) If True, multiply result by `gamma`.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Method Detail

axis

public LayerNorm.Builder axis(int... axis)

List the axis over which the mean and variance will be calculated (alternative to normalizedShape).

Parameters:

axis - input axis over which the mean and variance will be calculated (if null all existing dimensions)

Returns:

this Builder

optCenter

public LayerNorm.Builder optCenter(boolean val)

If True, add offset of `beta` to normalized tensor. Defaults to True.

Parameters:

val - True or False on whether to add and train offset value

Returns:

this Builder

optScale

public LayerNorm.Builder optScale(boolean val)

If True, multiply result by `gamma`. Defaults to True;

Parameters:

val - True or False on whether to add and train scale value

Returns:

this Builder

optEpsilon

public LayerNorm.Builder optEpsilon(float val)

Sets the epsilon value to prevent division by 0.

Parameters:

val - the epsilon value

Returns:

this Builder

build

public LayerNorm build()

Builds a LayerNorm block.

Returns:

the LayerNorm block