Package io.github.kirstenali.deepj.tensor.cpu

Class CpuBackend

java.lang.Object

io.github.kirstenali.deepj.tensor.cpu.CpuBackend

All Implemented Interfaces:

TensorBackend

public final class CpuBackend
extends Object

Constructor Summary

Constructors

Constructor	Description
CpuBackend()

Method Summary

Modifier and Type	Method	Description
void	adamWUpdate(Tensor w, Tensor g, Tensor mt, Tensor vt, float lr, float beta1, float beta2, float eps, float weightDecay, float bc1, float bc2)
Tensor	add(Tensor a, Tensor b)
Tensor	addBroadcastCols(Tensor a, Tensor cv)
Tensor	addBroadcastRows(Tensor a, Tensor rv)
void	addInPlace(Tensor a, Tensor b)
Tensor	addRowVector(Tensor a, Tensor rv)
Tensor	addScalar(Tensor a, float s)
void	addScalarInPlace(Tensor a, float s)
Tensor	batchedMatmul(Tensor left, Tensor right, int batches, boolean ta, boolean tb)
Tensor	cast(Tensor input, DType dtype)
Tensor	causalMask(int size)
Tensor	causalSoftmax(Tensor input, int length, float scale)
Tensor	clamp(Tensor a, float min, float max)
void	copyInto(Tensor target, Tensor source)	Explicitly assigns logical values, casting to the destination's storage dtype.
CrossEntropyResult	crossEntropy(Tensor logits, int[] targets)
CrossEntropyResult	crossEntropy(Tensor logits, int[] targets, boolean[] mask)
Tensor	crossEntropyGradient(Tensor logits, int[] targets)
Tensor	crossEntropyGradient(Tensor logits, int[] targets, boolean[] mask)
float	crossEntropyLoss(Tensor logits, int[] targets)
float	crossEntropyLoss(Tensor logits, int[] targets, boolean[] mask)
Tensor	divide(Tensor a, Tensor b)
Tensor	divideBroadcastCols(Tensor a, Tensor cv)
void	divideInPlace(Tensor a, Tensor b)
Tensor	divideScalar(Tensor a, float s)
void	divideScalarInPlace(Tensor a, float s)
Tensor	exp(Tensor a)
void	expInPlace(Tensor a)
Tensor	gelu(Tensor a)
Tensor	geluBackward(Tensor input, Tensor gradOutput)
void	geluInPlace(Tensor a)
float	get(Tensor t, int r, int c)
Tensor	getRow(Tensor t, int row)
void	indexAdd(Tensor target, int axis, int[] indices, Tensor updates)
Tensor	indexSelect(Tensor input, int axis, int[] indices)
Tensor	indexSelectGradient(Tensor gradient, int[] shape, int axis, int[] indices)
Tensor	layerNormBackward(Tensor dXHat, Tensor xHat, Tensor std, int dim)
Tensor	log(Tensor a)
void	logInPlace(Tensor a)
Tensor	matmul(Tensor a, Tensor b)
Tensor	maxAlongRows(Tensor a)
Tensor	maximumAxes(Tensor input, int[] axes, boolean keep)
Tensor	maximumAxesBackward(Tensor input, Tensor maximum, Tensor gradient, int[] axes, boolean keep)
Tensor	meanAlongRows(Tensor a)
Tensor	meanAxes(Tensor input, int[] axes, boolean keepDimensions)
Tensor	mergeHeads(Tensor input, int heads)
Tensor	multiply(Tensor a, Tensor b)
Tensor	multiplyBroadcastCols(Tensor a, Tensor cv)
Tensor	multiplyBroadcastRows(Tensor a, Tensor rv)
void	multiplyInPlace(Tensor a, Tensor b)
Tensor	multiplyScalar(Tensor a, float s)
void	multiplyScalarInPlace(Tensor a, float s)
Tensor	neg(Tensor a)
void	negInPlace(Tensor a)
Tensor	normalizationBackward(Tensor gradient, Tensor normalized, Tensor denominator, Tensor gamma, boolean centered)
NormalizationResult	normalize(Tensor input, Tensor gamma, Tensor beta, float epsilon, boolean centered)
Tensor	ones(int rows, int cols)
Tensor	pow(Tensor a, float exponent)
void	print(Tensor t, String label)
Tensor	random(int rows, int cols, Random rand)
Tensor	relu(Tensor a)
Tensor	reluBackward(Tensor input, Tensor gradOutput)
void	reluInPlace(Tensor a)
RmsNormResult	rmsNorm(Tensor input, Tensor gamma, float epsilon)
Tensor	rmsNormBackward(Tensor gradient, Tensor normalized, Tensor rms, Tensor gamma)
Tensor	sampleRows(Tensor t, int n, Random rnd)
void	scatterAddRows(Tensor target, int[] indices, Tensor grad)
void	set(Tensor t, int r, int c, float v)
void	setRow(Tensor t, int row, Tensor source, int srcRow)
Tensor	sigmoid(Tensor a)
void	sigmoidInPlace(Tensor a)
Tensor	sliceGradient(Tensor gradient, int[] shape, int axis, int start, int step)
Tensor	sliceRows(Tensor input, int[] rows)
Tensor	sliceRows(Tensor t, int[] rowIndices, int cols)
Tensor	softmaxAxis(Tensor input, int axis)
Tensor	softmaxBackward(Tensor gradOutput, Tensor softmaxOut)
Tensor	softmaxRows(Tensor logits)
Tensor	splitHeads(Tensor input, int heads)
Tensor	sqrt(Tensor a)
void	sqrtInPlace(Tensor a)
Tensor	subtract(Tensor a, Tensor b)
Tensor	subtractBroadcastCols(Tensor a, Tensor cv)
void	subtractInPlace(Tensor a, Tensor b)
float	sum(Tensor a)
float	sumAbs(Tensor a)
Tensor	sumAlongCols(Tensor a)
Tensor	sumAlongRows(Tensor a)
Tensor	sumAxes(Tensor input, int[] axes, boolean keepDimensions)
Tensor	sumRows(Tensor a)
Tensor	swiGlu(Tensor gate, Tensor up)
SwiGluBackwardResult	swiGluBackward(Tensor gradient, Tensor gate, Tensor up)
Tensor	tanh(Tensor a)
void	tanhInPlace(Tensor a)
Tensor	transpose(Tensor a)
Tensor	varianceAlongRows(Tensor a)
Tensor	varianceAxes(Tensor input, int[] axes, boolean keep, int correction)
Tensor	zeros(int rows, int cols)	Backend-owned positive-zero storage; ordinary Tensor constructors stay CPU-backed.
Tensor	zeros(DType dtype, int... shape)

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.tensor.TensorBackend

causalMask, copy, expand, headPrefixMatmul, l2Norm, materializeTensor, permute, releaseResources, releaseTemporaryResources, reshape, rotary, select, slice, softmaxAxisBackward, softmaxBackward, sumToShape, unsqueeze, usesDeviceStorage, view, writeHeadCache

Constructor Details

CpuBackend

public CpuBackend()

Method Details

causalSoftmax

public Tensor causalSoftmax(Tensor input,
 int length,
 float scale)

sliceRows

public Tensor sliceRows(Tensor input,
 int[] rows)

splitHeads

public Tensor splitHeads(Tensor input,
 int heads)

mergeHeads

public Tensor mergeHeads(Tensor input,
 int heads)

batchedMatmul

public Tensor batchedMatmul(Tensor left,
 Tensor right,
 int batches,
 boolean ta,
 boolean tb)

copyInto

public void copyInto(Tensor target,
 Tensor source)

Description copied from interface: TensorBackend

Explicitly assigns logical values, casting to the destination's storage dtype.

zeros

public Tensor zeros(DType dtype,
 int... shape)

cast

public Tensor cast(Tensor input,
 DType dtype)

sumAxes

public Tensor sumAxes(Tensor input,
 int[] axes,
 boolean keepDimensions)

meanAxes

public Tensor meanAxes(Tensor input,
 int[] axes,
 boolean keepDimensions)

varianceAxes

public Tensor varianceAxes(Tensor input,
 int[] axes,
 boolean keep,
 int correction)

maximumAxes

public Tensor maximumAxes(Tensor input,
 int[] axes,
 boolean keep)

maximumAxesBackward

public Tensor maximumAxesBackward(Tensor input,
 Tensor maximum,
 Tensor gradient,
 int[] axes,
 boolean keep)

sliceGradient

public Tensor sliceGradient(Tensor gradient,
 int[] shape,
 int axis,
 int start,
 int step)

softmaxAxis

public Tensor softmaxAxis(Tensor input,
 int axis)

indexSelect

public Tensor indexSelect(Tensor input,
 int axis,
 int[] indices)

indexAdd

public void indexAdd(Tensor target,
 int axis,
 int[] indices,
 Tensor updates)

indexSelectGradient

public Tensor indexSelectGradient(Tensor gradient,
 int[] shape,
 int axis,
 int[] indices)

crossEntropy

public CrossEntropyResult crossEntropy(Tensor logits,
 int[] targets)

crossEntropy

public CrossEntropyResult crossEntropy(Tensor logits,
 int[] targets,
 boolean[] mask)

normalize

public NormalizationResult normalize(Tensor input,
 Tensor gamma,
 Tensor beta,
 float epsilon,
 boolean centered)

normalizationBackward

public Tensor normalizationBackward(Tensor gradient,
 Tensor normalized,
 Tensor denominator,
 Tensor gamma,
 boolean centered)

rmsNorm

public RmsNormResult rmsNorm(Tensor input,
 Tensor gamma,
 float epsilon)

rmsNormBackward

public Tensor rmsNormBackward(Tensor gradient,
 Tensor normalized,
 Tensor rms,
 Tensor gamma)

swiGlu

public Tensor swiGlu(Tensor gate,
 Tensor up)

swiGluBackward

public SwiGluBackwardResult swiGluBackward(Tensor gradient,
 Tensor gate,
 Tensor up)

crossEntropyLoss

public float crossEntropyLoss(Tensor logits,
 int[] targets)

crossEntropyGradient

public Tensor crossEntropyGradient(Tensor logits,
 int[] targets)

crossEntropyLoss

public float crossEntropyLoss(Tensor logits,
 int[] targets,
 boolean[] mask)

crossEntropyGradient

public Tensor crossEntropyGradient(Tensor logits,
 int[] targets,
 boolean[] mask)

adamWUpdate

public void adamWUpdate(Tensor w,
 Tensor g,
 Tensor mt,
 Tensor vt,
 float lr,
 float beta1,
 float beta2,
 float eps,
 float weightDecay,
 float bc1,
 float bc2)

layerNormBackward

public Tensor layerNormBackward(Tensor dXHat,
 Tensor xHat,
 Tensor std,
 int dim)

get

public float get(Tensor t,
 int r,
 int c)

set

public void set(Tensor t,
 int r,
 int c,
 float v)

getRow

public Tensor getRow(Tensor t,
 int row)

setRow

public void setRow(Tensor t,
 int row,
 Tensor source,
 int srcRow)

sliceRows

public Tensor sliceRows(Tensor t,
 int[] rowIndices,
 int cols)

scatterAddRows

public void scatterAddRows(Tensor target,
 int[] indices,
 Tensor grad)

sampleRows

public Tensor sampleRows(Tensor t,
 int n,
 Random rnd)

addInPlace

public void addInPlace(Tensor a,
 Tensor b)

subtractInPlace

public void subtractInPlace(Tensor a,
 Tensor b)

multiplyInPlace

public void multiplyInPlace(Tensor a,
 Tensor b)

divideInPlace

public void divideInPlace(Tensor a,
 Tensor b)

multiplyScalarInPlace

public void multiplyScalarInPlace(Tensor a,
 float s)

addScalarInPlace

public void addScalarInPlace(Tensor a,
 float s)

divideScalarInPlace

public void divideScalarInPlace(Tensor a,
 float s)

sqrtInPlace

public void sqrtInPlace(Tensor a)

negInPlace

public void negInPlace(Tensor a)

expInPlace

public void expInPlace(Tensor a)

logInPlace

public void logInPlace(Tensor a)

reluInPlace

public void reluInPlace(Tensor a)

geluInPlace

public void geluInPlace(Tensor a)

tanhInPlace

public void tanhInPlace(Tensor a)

sigmoidInPlace

public void sigmoidInPlace(Tensor a)

print

public void print(Tensor t,
 String label)

sumRows

public Tensor sumRows(Tensor a)

sumAlongRows

public Tensor sumAlongRows(Tensor a)

sumAlongCols

public Tensor sumAlongCols(Tensor a)

meanAlongRows

public Tensor meanAlongRows(Tensor a)

varianceAlongRows

public Tensor varianceAlongRows(Tensor a)

maxAlongRows

public Tensor maxAlongRows(Tensor a)

sum

public float sum(Tensor a)

sumAbs

public float sumAbs(Tensor a)

transpose

public Tensor transpose(Tensor a)

clamp

public Tensor clamp(Tensor a,
 float min,
 float max)

sqrt

public Tensor sqrt(Tensor a)

pow

public Tensor pow(Tensor a,
 float exponent)

neg

public Tensor neg(Tensor a)

exp

public Tensor exp(Tensor a)

log

public Tensor log(Tensor a)

tanh

public Tensor tanh(Tensor a)

sigmoid

public Tensor sigmoid(Tensor a)

relu

public Tensor relu(Tensor a)

gelu

public Tensor gelu(Tensor a)

reluBackward

public Tensor reluBackward(Tensor input,
 Tensor gradOutput)

geluBackward

public Tensor geluBackward(Tensor input,
 Tensor gradOutput)

softmaxRows

public Tensor softmaxRows(Tensor logits)

softmaxBackward

public Tensor softmaxBackward(Tensor gradOutput,
 Tensor softmaxOut)

zeros

public Tensor zeros(int rows,
 int cols)

Description copied from interface: TensorBackend

Backend-owned positive-zero storage; ordinary Tensor constructors stay CPU-backed.

Specified by:

zeros in interface TensorBackend

ones

public Tensor ones(int rows,
 int cols)

random

public Tensor random(int rows,
 int cols,
 Random rand)

causalMask

public Tensor causalMask(int size)

matmul

public Tensor matmul(Tensor a,
 Tensor b)

Specified by:

matmul in interface TensorBackend

add

public Tensor add(Tensor a,
 Tensor b)

Specified by:

add in interface TensorBackend

subtract

public Tensor subtract(Tensor a,
 Tensor b)

Specified by:

subtract in interface TensorBackend

multiply

public Tensor multiply(Tensor a,
 Tensor b)

Specified by:

multiply in interface TensorBackend

divide

public Tensor divide(Tensor a,
 Tensor b)

Specified by:

divide in interface TensorBackend

addRowVector

public Tensor addRowVector(Tensor a,
 Tensor rv)

Specified by:

addRowVector in interface TensorBackend

addBroadcastRows

public Tensor addBroadcastRows(Tensor a,
 Tensor rv)

Specified by:

addBroadcastRows in interface TensorBackend

multiplyBroadcastRows

public Tensor multiplyBroadcastRows(Tensor a,
 Tensor rv)

Specified by:

multiplyBroadcastRows in interface TensorBackend

addBroadcastCols

public Tensor addBroadcastCols(Tensor a,
 Tensor cv)

Specified by:

addBroadcastCols in interface TensorBackend

subtractBroadcastCols

public Tensor subtractBroadcastCols(Tensor a,
 Tensor cv)

Specified by:

subtractBroadcastCols in interface TensorBackend

multiplyBroadcastCols

public Tensor multiplyBroadcastCols(Tensor a,
 Tensor cv)

Specified by:

multiplyBroadcastCols in interface TensorBackend

divideBroadcastCols

public Tensor divideBroadcastCols(Tensor a,
 Tensor cv)

Specified by:

divideBroadcastCols in interface TensorBackend

multiplyScalar

public Tensor multiplyScalar(Tensor a,
 float s)

Specified by:

multiplyScalar in interface TensorBackend

addScalar

public Tensor addScalar(Tensor a,
 float s)

Specified by:

addScalar in interface TensorBackend

divideScalar

public Tensor divideScalar(Tensor a,
 float s)

Specified by:

divideScalar in interface TensorBackend