Package io.github.kirstenali.deepj.transformer.embeddings

Class Embedding

java.lang.Object

io.github.kirstenali.deepj.transformer.embeddings.Embedding

All Implemented Interfaces:

Trainable

public final class Embedding
extends Object
implements Trainable

Token embedding: ids -> vectors. Input ids are provided via forward(int[]). Output is a tensor of shape [nTokens x dModel].

Constructor Summary

Constructors

Constructor	Description
Embedding(int vocabSize, int dModel, Random rnd)

Method Summary

Modifier and Type	Method	Description
void	backward(Tensor gradOut)
Tensor	forward(int[] ids)
List<Parameter>	parameters()
Parameter	weight()

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.training.Trainable

zeroGrad

Constructor Details

Embedding

public Embedding(int vocabSize,
 int dModel,
 Random rnd)

Method Details

forward

public Tensor forward(int[] ids)

backward

public void backward(Tensor gradOut)

weight

public Parameter weight()

parameters

public List<Parameter> parameters()

Specified by:

parameters in interface Trainable