Package me.damoebe.architectures.transformer

Class Block<H extends Head,N extends FFNetwork>

java.lang.Object

me.damoebe.architectures.transformer.Block<H,N>

Type Parameters:

H - The Head type that should be used to construct the MultiHeadAttention attribute.

N - The MLP type that will be used in this block.

public class Block<H extends Head,N extends FFNetwork>
extends Object

The Transformer Bock class containing a mha and multiple mlps.

Constructor Summary

Constructors

Constructor	Description
Block(@NotNull MultiHeadAttention<H> multiHeadAttention, N multiLayerPerceptron)	The main constructor of the Block class

Method Summary

Modifier and Type	Method	Description
Block<H,N>	clone()
Sequence	getOutputFor(Sequence... inputEmbeddings)	Gets the output for the inserted input(s) by passing them through the mha module and mlp.

Methods inherited from class java.lang.Object

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

Constructor Details

Block

public Block(@NotNull
 @NotNull MultiHeadAttention<H> multiHeadAttention,
 @NotNull
 N multiLayerPerceptron)

The main constructor of the Block class

Parameters:

multiHeadAttention - A MultiHeadAttention object that will be cloned and used for this block.

multiLayerPerceptron - A FFNetwork object that will be cloned and used as an MLP architecture.

Method Details

getOutputFor

public Sequence getOutputFor(Sequence... inputEmbeddings)

Gets the output for the inserted input(s) by passing them through the mha module and mlp.

Parameters:

inputEmbeddings - A list of inputs: if the head type is EDHead: [0] is the decoder input and [1] is the encoder input. Else: [0] is the head input.

Returns:

The output embeddings for this block.

clone

public Block<H,N> clone()

Overrides:

clone in class Object