Package io.github.kirstenali.deepj.nn.transformer

Class DecoderOnlyModel

java.lang.Object

io.github.kirstenali.deepj.nn.transformer.DecoderOnlyModel

All Implemented Interfaces:

CausalLM, Persistable, Trainable

Direct Known Subclasses:

DeepJLoop, DeepJNova, DeepJOrbit, DeepJOrigin, DeepJPrism

public abstract class DecoderOnlyModel
extends Object
implements CausalLM, Persistable

Field Summary

Fields

Modifier and Type	Field	Description
protected final Projection	lmHead
protected final NormLayer	normF
protected final TransformerStack	stack
protected final Embedding	tokEmb

Constructor Summary

Constructors

Modifier	Constructor	Description
protected	DecoderOnlyModel(Embedding tokEmb, TransformerStack stack, NormLayer normF, Projection lmHead)

Method Summary

Modifier and Type	Method	Description
protected final void	applyInitScale(float factor)
abstract TransformerConfig	config()
protected Tensor	embed(int[] inputIds)
protected Tensor	embed(int[] inputIds, int offset)
protected Tensor	embedBatch(int[][] inputIds)
protected List<Parameter>	embeddingParameters()
Tensor	forward(int[] inputIds)
Tensor	forwardBatch(int[][] inputIds)
final void	load(Path path)
final Map<String,String>	metadata()
GenerationSession	openGeneration()
GenerationSession	openGeneration(ContextWindowMode mode)
List<Parameter>	parameters()
final PromptFormat	promptFormat()
final void	save(Path path)
final void	setMetadata(Map<String,String> values)
final void	setPromptFormat(PromptFormat format)
final boolean	supportsBatchedForward()
protected boolean	supportsCachedGeneration()
protected boolean	supportsRollingGeneration()

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.nn.transformer.CausalLM

gradClipNorm

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

zeroGrad

Field Details

tokEmb

protected final Embedding tokEmb

stack

protected final TransformerStack stack

normF

protected final NormLayer normF

lmHead

protected final Projection lmHead

Constructor Details

DecoderOnlyModel

protected DecoderOnlyModel(Embedding tokEmb,
 TransformerStack stack,
 NormLayer normF,
 Projection lmHead)

Method Details

config

public abstract TransformerConfig config()

promptFormat

public final PromptFormat promptFormat()

setPromptFormat

public final void setPromptFormat(PromptFormat format)

metadata

public final Map<String,String> metadata()

setMetadata

public final void setMetadata(Map<String,String> values)

save

public final void save(Path path)
                throws IOException

Specified by:

save in interface Persistable

Throws:

IOException

load

public final void load(Path path)
                throws IOException

Specified by:

load in interface Persistable

Throws:

IOException

embed

protected Tensor embed(int[] inputIds)

embed

protected Tensor embed(int[] inputIds,
 int offset)

embedBatch

protected Tensor embedBatch(int[][] inputIds)

supportsCachedGeneration

protected boolean supportsCachedGeneration()

supportsRollingGeneration

protected boolean supportsRollingGeneration()

openGeneration

public GenerationSession openGeneration()

Specified by:

openGeneration in interface CausalLM

openGeneration

public GenerationSession openGeneration(ContextWindowMode mode)

Specified by:

openGeneration in interface CausalLM

embeddingParameters

protected List<Parameter> embeddingParameters()

forward

public Tensor forward(int[] inputIds)

Specified by:

forward in interface CausalLM

supportsBatchedForward

public final boolean supportsBatchedForward()

Specified by:

supportsBatchedForward in interface CausalLM

forwardBatch

public Tensor forwardBatch(int[][] inputIds)

Specified by:

forwardBatch in interface CausalLM

parameters

public List<Parameter> parameters()

Specified by:

parameters in interface Persistable

Specified by:

parameters in interface Trainable

applyInitScale

protected final void applyInitScale(float factor)