GPU
CPU
NPU
WebNN Operations · FP16 · GPU CPU NPU
GPU
preferredInputLayout
maxTensorByteLength
CPU
preferredInputLayout
maxTensorByteLength
NPU
preferredInputLayout
maxTensorByteLength
GPU
f16
f32
i32
i4
i64
i8
u32
u4
u64
u8
min
max
abs
input
abs
output
add
a
add
b
add
output
argMax
input
argMax
output
argMin
input
argMin
output
averagePool2d
input
averagePool2d
output
batchNormalization
bias
batchNormalization
input
batchNormalization
mean
batchNormalization
output
batchNormalization
scale
batchNormalization
variance
cast
input
cast
output
ceil
input
ceil
output
clamp
input
clamp
output
concat
inputs
concat
output
conv2d
bias
conv2d
filter
conv2d
input
conv2d
output
convTranspose2d
bias
convTranspose2d
filter
convTranspose2d
input
convTranspose2d
output
cos
input
cos
output
cumulativeSum
input
cumulativeSum
output
dequantizeLinear
input
dequantizeLinear
output
dequantizeLinear
scale
dequantizeLinear
zeroPoint
div
a
div
b
div
output
elu
input
elu
output
equal
a
equal
b
equal
output
erf
input
erf
output
exp
input
exp
output
expand
input
expand
output
floor
input
floor
output
gather
indices
gather
input
gather
output
gatherElements
indices
gatherElements
input
gatherElements
output
gatherND
indices
gatherND
input
gatherND
output
gelu
input
gelu
output
gemm
a
gemm
b
gemm
c
gemm
output
greater
a
greater
b
greater
output
greaterOrEqual
a
greaterOrEqual
b
greaterOrEqual
output
gru
bias
gru
initialHiddenState
gru
input
gru
output0
gru
output1
gru
recurrentBias
gru
recurrentWeight
gru
weight
gruCell
bias
gruCell
hiddenState
gruCell
input
gruCell
output
gruCell
recurrentBias
gruCell
recurrentWeight
gruCell
weight
hardSigmoid
input
hardSigmoid
output
hardSwish
input
hardSwish
output
identity
input
identity
output
instanceNormalization
bias
instanceNormalization
input
instanceNormalization
output
instanceNormalization
scale
isInfinite
a
isInfinite
output
isNaN
a
isNaN
output
l2Pool2d
input
l2Pool2d
output
layerNormalization
bias
layerNormalization
input
layerNormalization
output
layerNormalization
scale
leakyRelu
input
leakyRelu
output
lesser
a
lesser
b
lesser
output
lesserOrEqual
a
lesserOrEqual
b
lesserOrEqual
output
linear
input
linear
output
log
input
log
output
logicalAnd
a
logicalAnd
b
logicalAnd
output
logicalNot
a
logicalNot
output
logicalOr
a
logicalOr
b
logicalOr
output
logicalXor
a
logicalXor
b
logicalXor
output
lstm
bias
lstm
initialCellState
lstm
initialHiddenState
lstm
input
lstm
output0
lstm
output1
lstm
output2
lstm
peepholeWeight
lstm
recurrentBias
lstm
recurrentWeight
lstm
weight
lstmCell
bias
lstmCell
cellState
lstmCell
hiddenState
lstmCell
input
lstmCell
output0
lstmCell
output1
lstmCell
peepholeWeight
lstmCell
recurrentBias
lstmCell
recurrentWeight
lstmCell
weight
matmul
a
matmul
b
matmul
output
max
a
max
b
max
output
maxPool2d
input
maxPool2d
output
min
a
min
b
min
output
mul
a
mul
b
mul
output
neg
input
neg
output
notEqual
a
notEqual
b
notEqual
output
pad
input
pad
output
pow
a
pow
b
pow
output
prelu
input
prelu
output
prelu
slope
quantizeLinear
input
quantizeLinear
output
quantizeLinear
scale
quantizeLinear
zeroPoint
reciprocal
input
reciprocal
output
reduceL1
input
reduceL1
output
reduceL2
input
reduceL2
output
reduceLogSum
input
reduceLogSum
output
reduceLogSumExp
input
reduceLogSumExp
output
reduceMax
input
reduceMax
output
reduceMean
input
reduceMean
output
reduceMin
input
reduceMin
output
reduceProduct
input
reduceProduct
output
reduceSum
input
reduceSum
output
reduceSumSquare
input
reduceSumSquare
output
relu
input
relu
output
resample2d
input
resample2d
output
reshape
input
reshape
output
reverse
input
reverse
output
roundEven
input
roundEven
output
scatterElements
indices
scatterElements
input
scatterElements
output
scatterElements
updates
scatterND
indices
scatterND
input
scatterND
output
scatterND
updates
sigmoid
input
sigmoid
output
sign
input
sign
output
sin
input
sin
output
slice
input
slice
output
softmax
input
softmax
output
softplus
input
softplus
output
softsign
input
softsign
output
split
input
split
outputs
sqrt
input
sqrt
output
sub
a
sub
b
sub
output
tan
input
tan
output
tanh
input
tanh
output
tile
input
tile
output
transpose
input
transpose
output
triangular
input
triangular
output
where
condition
where
falseValue
where
output
where
trueValue
Allowed data types in WebNN Spec
CPU
f16
f32
i32
i4
i64
i8
u32
u4
u64
u8
min
max
abs
input
abs
output
add
a
add
b
add
output
argMax
input
argMax
output
argMin
input
argMin
output
averagePool2d
input
averagePool2d
output
batchNormalization
bias
batchNormalization
input
batchNormalization
mean
batchNormalization
output
batchNormalization
scale
batchNormalization
variance
cast
input
cast
output
ceil
input
ceil
output
clamp
input
clamp
output
concat
inputs
concat
output
conv2d
bias
conv2d
filter
conv2d
input
conv2d
output
convTranspose2d
bias
convTranspose2d
filter
convTranspose2d
input
convTranspose2d
output
cos
input
cos
output
cumulativeSum
input
cumulativeSum
output
dequantizeLinear
input
dequantizeLinear
output
dequantizeLinear
scale
dequantizeLinear
zeroPoint
div
a
div
b
div
output
elu
input
elu
output
equal
a
equal
b
equal
output
erf
input
erf
output
exp
input
exp
output
expand
input
expand
output
floor
input
floor
output
gather
indices
gather
input
gather
output
gatherElements
indices
gatherElements
input
gatherElements
output
gatherND
indices
gatherND
input
gatherND
output
gelu
input
gelu
output
gemm
a
gemm
b
gemm
c
gemm
output
greater
a
greater
b
greater
output
greaterOrEqual
a
greaterOrEqual
b
greaterOrEqual
output
gru
bias
gru
initialHiddenState
gru
input
gru
output0
gru
output1
gru
recurrentBias
gru
recurrentWeight
gru
weight
gruCell
bias
gruCell
hiddenState
gruCell
input
gruCell
output
gruCell
recurrentBias
gruCell
recurrentWeight
gruCell
weight
hardSigmoid
input
hardSigmoid
output
hardSwish
input
hardSwish
output
identity
input
identity
output
instanceNormalization
bias
instanceNormalization
input
instanceNormalization
output
instanceNormalization
scale
isInfinite
a
isInfinite
output
isNaN
a
isNaN
output
l2Pool2d
input
l2Pool2d
output
layerNormalization
bias
layerNormalization
input
layerNormalization
output
layerNormalization
scale
leakyRelu
input
leakyRelu
output
lesser
a
lesser
b
lesser
output
lesserOrEqual
a
lesserOrEqual
b
lesserOrEqual
output
linear
input
linear
output
log
input
log
output
logicalAnd
a
logicalAnd
b
logicalAnd
output
logicalNot
a
logicalNot
output
logicalOr
a
logicalOr
b
logicalOr
output
logicalXor
a
logicalXor
b
logicalXor
output
lstm
bias
lstm
initialCellState
lstm
initialHiddenState
lstm
input
lstm
output0
lstm
output1
lstm
output2
lstm
peepholeWeight
lstm
recurrentBias
lstm
recurrentWeight
lstm
weight
lstmCell
bias
lstmCell
cellState
lstmCell
hiddenState
lstmCell
input
lstmCell
output0
lstmCell
output1
lstmCell
peepholeWeight
lstmCell
recurrentBias
lstmCell
recurrentWeight
lstmCell
weight
matmul
a
matmul
b
matmul
output
max
a
max
b
max
output
maxPool2d
input
maxPool2d
output
min
a
min
b
min
output
mul
a
mul
b
mul
output
neg
input
neg
output
notEqual
a
notEqual
b
notEqual
output
pad
input
pad
output
pow
a
pow
b
pow
output
prelu
input
prelu
output
prelu
slope
quantizeLinear
input
quantizeLinear
output
quantizeLinear
scale
quantizeLinear
zeroPoint
reciprocal
input
reciprocal
output
reduceL1
input
reduceL1
output
reduceL2
input
reduceL2
output
reduceLogSum
input
reduceLogSum
output
reduceLogSumExp
input
reduceLogSumExp
output
reduceMax
input
reduceMax
output
reduceMean
input
reduceMean
output
reduceMin
input
reduceMin
output
reduceProduct
input
reduceProduct
output
reduceSum
input
reduceSum
output
reduceSumSquare
input
reduceSumSquare
output
relu
input
relu
output
resample2d
input
resample2d
output
reshape
input
reshape
output
reverse
input
reverse
output
roundEven
input
roundEven
output
scatterElements
indices
scatterElements
input
scatterElements
output
scatterElements
updates
scatterND
indices
scatterND
input
scatterND
output
scatterND
updates
sigmoid
input
sigmoid
output
sign
input
sign
output
sin
input
sin
output
slice
input
slice
output
softmax
input
softmax
output
softplus
input
softplus
output
softsign
input
softsign
output
split
input
split
outputs
sqrt
input
sqrt
output
sub
a
sub
b
sub
output
tan
input
tan
output
tanh
input
tanh
output
tile
input
tile
output
transpose
input
transpose
output
triangular
input
triangular
output
where
condition
where
falseValue
where
output
where
trueValue
Allowed data types in WebNN Spec
NPU
f16
f32
i32
i4
i64
i8
u32
u4
u64
u8
min
max
abs
input
abs
output
add
a
add
b
add
output
argMax
input
argMax
output
argMin
input
argMin
output
averagePool2d
input
averagePool2d
output
batchNormalization
bias
batchNormalization
input
batchNormalization
mean
batchNormalization
output
batchNormalization
scale
batchNormalization
variance
cast
input
cast
output
ceil
input
ceil
output
clamp
input
clamp
output
concat
inputs
concat
output
conv2d
bias
conv2d
filter
conv2d
input
conv2d
output
convTranspose2d
bias
convTranspose2d
filter
convTranspose2d
input
convTranspose2d
output
cos
input
cos
output
cumulativeSum
input
cumulativeSum
output
dequantizeLinear
input
dequantizeLinear
output
dequantizeLinear
scale
dequantizeLinear
zeroPoint
div
a
div
b
div
output
elu
input
elu
output
equal
a
equal
b
equal
output
erf
input
erf
output
exp
input
exp
output
expand
input
expand
output
floor
input
floor
output
gather
indices
gather
input
gather
output
gatherElements
indices
gatherElements
input
gatherElements
output
gatherND
indices
gatherND
input
gatherND
output
gelu
input
gelu
output
gemm
a
gemm
b
gemm
c
gemm
output
greater
a
greater
b
greater
output
greaterOrEqual
a
greaterOrEqual
b
greaterOrEqual
output
gru
bias
gru
initialHiddenState
gru
input
gru
output0
gru
output1
gru
recurrentBias
gru
recurrentWeight
gru
weight
gruCell
bias
gruCell
hiddenState
gruCell
input
gruCell
output
gruCell
recurrentBias
gruCell
recurrentWeight
gruCell
weight
hardSigmoid
input
hardSigmoid
output
hardSwish
input
hardSwish
output
identity
input
identity
output
instanceNormalization
bias
instanceNormalization
input
instanceNormalization
output
instanceNormalization
scale
isInfinite
a
isInfinite
output
isNaN
a
isNaN
output
l2Pool2d
input
l2Pool2d
output
layerNormalization
bias
layerNormalization
input
layerNormalization
output
layerNormalization
scale
leakyRelu
input
leakyRelu
output
lesser
a
lesser
b
lesser
output
lesserOrEqual
a
lesserOrEqual
b
lesserOrEqual
output
linear
input
linear
output
log
input
log
output
logicalAnd
a
logicalAnd
b
logicalAnd
output
logicalNot
a
logicalNot
output
logicalOr
a
logicalOr
b
logicalOr
output
logicalXor
a
logicalXor
b
logicalXor
output
lstm
bias
lstm
initialCellState
lstm
initialHiddenState
lstm
input
lstm
output0
lstm
output1
lstm
output2
lstm
peepholeWeight
lstm
recurrentBias
lstm
recurrentWeight
lstm
weight
lstmCell
bias
lstmCell
cellState
lstmCell
hiddenState
lstmCell
input
lstmCell
output0
lstmCell
output1
lstmCell
peepholeWeight
lstmCell
recurrentBias
lstmCell
recurrentWeight
lstmCell
weight
matmul
a
matmul
b
matmul
output
max
a
max
b
max
output
maxPool2d
input
maxPool2d
output
min
a
min
b
min
output
mul
a
mul
b
mul
output
neg
input
neg
output
notEqual
a
notEqual
b
notEqual
output
pad
input
pad
output
pow
a
pow
b
pow
output
prelu
input
prelu
output
prelu
slope
quantizeLinear
input
quantizeLinear
output
quantizeLinear
scale
quantizeLinear
zeroPoint
reciprocal
input
reciprocal
output
reduceL1
input
reduceL1
output
reduceL2
input
reduceL2
output
reduceLogSum
input
reduceLogSum
output
reduceLogSumExp
input
reduceLogSumExp
output
reduceMax
input
reduceMax
output
reduceMean
input
reduceMean
output
reduceMin
input
reduceMin
output
reduceProduct
input
reduceProduct
output
reduceSum
input
reduceSum
output
reduceSumSquare
input
reduceSumSquare
output
relu
input
relu
output
resample2d
input
resample2d
output
reshape
input
reshape
output
reverse
input
reverse
output
roundEven
input
roundEven
output
scatterElements
indices
scatterElements
input
scatterElements
output
scatterElements
updates
scatterND
indices
scatterND
input
scatterND
output
scatterND
updates
sigmoid
input
sigmoid
output
sign
input
sign
output
sin
input
sin
output
slice
input
slice
output
softmax
input
softmax
output
softplus
input
softplus
output
softsign
input
softsign
output
split
input
split
outputs
sqrt
input
sqrt
output
sub
a
sub
b
sub
output
tan
input
tan
output
tanh
input
tanh
output
tile
input
tile
output
transpose
input
transpose
output
triangular
input
triangular
output
where
condition
where
falseValue
where
output
where
trueValue
Allowed data types in WebNN Spec