Package me.damoebe.architectures.transformer

Class DecoderTransformer<N extends FFNetwork>

java.lang.Object

me.damoebe.architectures.Network

me.damoebe.architectures.transformer.DecoderTransformer<N>

Type Parameters:

N - The feed-forward neural network type used for this transformer

public class DecoderTransformer<N extends FFNetwork>
extends Network

The class for a Decoder-Only Transformer, used for generative tasks

Constructor Summary

Constructors

Constructor	Description
DecoderTransformer(int decoderBlockAmount, Block<Head,N> referenceBlock)	Main constructor for a Decoder-Only Transformer with a reference Decoder-Block

Method Summary

Modifier and Type	Method	Description
List<Double>	getOutput()
void	insertInput(List<Double> input)
void	train(List<Double> input, List<Double> optimalOutput)
void	updateAllActivations()	Updates All activations in the transformer as well as the currentOutput

Methods inherited from class me.damoebe.architectures.Network

loadNetworkFromJson, loadToJsonFile

Methods inherited from class java.lang.Object

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

Constructor Details

DecoderTransformer

public DecoderTransformer(int decoderBlockAmount,
 Block<Head,N> referenceBlock)

Main constructor for a Decoder-Only Transformer with a reference Decoder-Block

Parameters:

decoderBlockAmount - The amount of Decoder Blocks that should be used.

referenceBlock - The reference Decoder-Block that will be used to construct the blocks

Method Details

train

public void train(List<Double> input,
 List<Double> optimalOutput)

Specified by:

train in class Network

getOutput

public List<Double> getOutput()

Specified by:

getOutput in class Network

insertInput

public void insertInput(List<Double> input)

Specified by:

insertInput in class Network

updateAllActivations

public void updateAllActivations()

Updates All activations in the transformer as well as the currentOutput

Specified by:

updateAllActivations in class Network