Package io.github.kirstenali.deepj.nn

Class EncoderAttention

java.lang.Object

io.github.kirstenali.deepj.nn.Module

io.github.kirstenali.deepj.nn.EncoderAttention

All Implemented Interfaces:

Layer, Trainable

public final class EncoderAttention
extends Module

Bidirectional multi-head attention with optional padding masks and backend-resident softmax.

Constructor Summary

Constructors

Constructor	Description
EncoderAttention(int width, int heads, float dropout, Random random, ParameterFactory parameters)

Method Summary

Modifier and Type	Method	Description
protected List<Layer>	children()
Tensor	forward(Tensor input)
Tensor	forward(Tensor input, boolean[][] validTokens)

Methods inherited from class io.github.kirstenali.deepj.nn.Module

buffers, eval, isTraining, parameters, setTraining, train

Methods inherited from class java.lang.Object

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

Methods inherited from interface io.github.kirstenali.deepj.layers.Layer

forward, supportsIncrementalInference

Methods inherited from interface io.github.kirstenali.deepj.training.Trainable

zeroGrad

Constructor Details

EncoderAttention

public EncoderAttention(int width,
 int heads,
 float dropout,
 Random random,
 ParameterFactory parameters)

Method Details

children

protected List<Layer> children()

Overrides:

children in class Module

forward

public Tensor forward(Tensor input)

forward

public Tensor forward(Tensor input,
 boolean[][] validTokens)