Package smile.llm.llama

Class TransformerBlock

java.lang.Object

smile.llm.llama.TransformerBlock

public class TransformerBlock
extends Object

A block in Transformer model. It consists of an attention mechanism followed by a feedforward neural network. This module can be stacked multiple times to create a complete Transformer model.

Constructor Summary

Constructors

Constructor	Description
TransformerBlock(int layerId, ModelArgs args)	Constructor.

Method Summary

Modifier and Type	Method	Description
Tensor	forward(Tensor x, int startPos, Tensor cis, Tensor mask)	Forward pass through the block.

Methods inherited from class java.lang.Object

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

Constructor Details

TransformerBlock

public TransformerBlock(int layerId,
 ModelArgs args)

Constructor.

Parameters:

layerId - the identifier of the block.

args - the model configuration parameters.

Method Details

forward

public Tensor forward(Tensor x,
 int startPos,
 Tensor cis,
 Tensor mask)

Forward pass through the block.

Parameters:

x - the input tensor.

startPos - the starting position for attention caching.

cis - the precomputed frequency tensor.

mask - the attention mask tensor.

Returns:

the output tensor.