ai.onnx#

This page lists all operators in the ai.onnx domain.

Operator

Since version

Deprecated

Short description

Abs

13

No

Absolute takes one input data (Tensor ) and produces one output data

Acos

22

No

Computes the Acos value of the input tensor element-wise.

Acosh

22

No

Computes the Acosh value of the input tensor element-wise.

Add

14

No

Performs element-wise binary addition (with Numpy-style broadcasting support).

AffineGrid

20

No

Generates a 2D or 3D flow field (sampling grid), given a batch of affine matr…

And

7

No

Returns the tensor resulted from performing the And logical operation

ArgMax

13

No

Computes the indices of the max elements of the input tensor’s element along …

ArgMin

13

No

Computes the indices of the min elements of the input tensor’s element along …

Asin

22

No

Computes the Asin value of the input tensor element-wise.

Asinh

22

No

Computes the Asinh value of the input tensor element-wise.

Atan

22

No

Computes the Atan value of the input tensor element-wise.

Atanh

22

No

Computes the Atanh value of the input tensor element-wise.

Attention

24

No

Computes scaled dot product attention on query, key and value tensors, using …

AveragePool

22

No

AveragePool consumes an input tensor X and applies average pooling across

BatchNormalization

15

No

Carries out batch normalization as described in the paper

Bernoulli

22

No

Draws binary random numbers (0 or 1) from a Bernoulli distribution. The input…

BitCast

26

No

Reinterprets the binary representation of a tensor as a different data type,

BitShift

11

No

Bitwise shift operator performs element-wise operation. For each input elemen…

BitwiseAnd

18

No

Returns the tensor resulting from performing the bitwise BitwiseAnd operation

BitwiseNot

18

No

Returns the bitwise not of the input tensor element-wise.

BitwiseOr

18

No

Returns the tensor resulting from performing the bitwise BitwiseOr operation

BitwiseXor

18

No

Returns the tensor resulting from performing the bitwise BitwiseXor operation

BlackmanWindow

17

No

Generates a Blackman window as described in the paper https://ieeexplore.ieee

Cast

25

No

The operator casts the elements of a given input tensor to a data type

CastLike

25

No

The operator casts the elements of a given input tensor (the first input) to

CausalConvWithState

27

No

Stateful causal 1D depthwise convolution.

Ceil

13

No

Ceil takes one input data (Tensor ) and produces one output data

Celu

28

No

Continuously Differentiable Exponential Linear Units:

CenterCropPad

18

No

Center crop or pad an input to given dimensions.

Clip

13

No

Clip operator limits the given input within an interval. The interval is

Col2Im

18

No

The operator rearranges column blocks back into a multidimensional image

Compress

11

No

Selects slices from an input tensor along a given axis value passed.

Concat

13

No

Concatenate a list of tensors into a single tensor. All input tensors must ha…

ConcatFromSequence

11

No

Concatenate a sequence of tensors into a single tensor.

Constant

25

No

This operator produces a constant tensor. Exactly one of the provided attribu…

ConstantOfShape

25

No

Generate a tensor with given value and shape.

Conv

22

No

The convolution operator consumes an input tensor and a filter, and

ConvInteger

10

No

The integer convolution operator consumes an input tensor, its zero-point, a …

ConvTranspose

22

No

The convolution transpose operator consumes an input tensor and a filter,

Cos

22

No

Computes the Cos value of the input tensor element-wise.

Cosh

22

No

Computes the Cosh value of the input tensor element-wise.

CumProd

26

No

Performs cumulative product of the input elements along the given axis.

CumSum

14

No

Performs cumulative sum of the input elements along the given axis.

DFT

20

No

Computes the discrete Fourier Transform (DFT) of the input.

DeformConv

22

No

Performs deformable convolution as described in https://arxiv.org/abs/1703.06

DepthToSpace

13

No

DepthToSpace rearranges (permutes) data from depth into blocks of spatial data.

DequantizeLinear

25

No

The linear dequantization operator. It consumes a quantized tensor, a scale, …

Det

22

No

Det calculates determinant of a square matrix or batches of square matrices.

Div

14

No

Performs element-wise binary division (with Numpy-style broadcasting support).

Dropout

22

No

DynamicQuantizeLinear

11

No

A Function to fuse calculation for Scale, Zero Point and FP32->8Bit conversio…

Einsum

12

No

An einsum of the form term1, term2 -> output-term produces an output tensor u…

Elu

22

No

Elu takes one input data (Tensor ) and produces one output data

Equal

19

No

Returns the tensor resulted from performing the Equal logical operation

Erf

13

No

Computes the Erf value of the input tensor element-wise.

Exp

13

No

Computes the Exp value of the input tensor element-wise.

Expand

13

No

Broadcast the input tensor following the given shape and the broadcast rule.

EyeLike

22

No

Generate a 2D tensor (matrix) with ones on the diagonal and zeros everywhere …

Flatten

25

No

Flattens the input tensor into a 2D matrix. If input tensor has shape

Floor

13

No

Floor takes one input data (Tensor ) and produces one output data

GRU

22

No

Computes an one-layer GRU. This operator is usually supported via some custom

Gather

13

No

Given data tensor of rank r >= 1, and indices tensor of rank q, gather

GatherElements

13

No

GatherElements takes two inputs data and indices of the same rank r >= 1

GatherND

13

No

Given data tensor of rank r >= 1, indices tensor of rank q >= 1, and batch_di…

Gelu

20

No

Gelu takes one input data (Tensor ) and produces one

Gemm

13

No

General Matrix multiplication:

GlobalAveragePool

22

No

GlobalAveragePool consumes an input tensor X and applies average pooling across

GlobalLpPool

22

No

GlobalLpPool consumes an input tensor X and applies lp pool pooling across

GlobalMaxPool

22

No

GlobalMaxPool consumes an input tensor X and applies max pooling across

Greater

13

No

Returns the tensor resulted from performing the Greater logical operation

GreaterOrEqual

16

No

Returns the tensor resulted from performing the GreaterOrEqual logical operation

GridSample

22

No

Given an input X and a flow-field grid, computes the output Y using X values …

GroupNormalization

21

No

A GroupNormalization function. Carries out group normalization as described in

HammingWindow

17

No

Generates a Hamming window as described in the paper https://ieeexplore.ieee….

HannWindow

17

No

Generates a Hann window as described in the paper https://ieeexplore.ieee.org

HardSigmoid

22

No

HardSigmoid takes one input data (Tensor ) and produces one output data

HardSwish

22

No

HardSwish takes one input data (Tensor ) and produces one output data (Tensor…

Hardmax

13

No

The operator computes the hardmax values for the given input.

Identity

25

No

Identity operator

If

13

No

If conditional

ImageDecoder

20

No

Loads and decodes and image from a file. If it can’t decode for any reason (e…

InstanceNormalization

22

No

Carries out instance normalization as described in the paper

IsInf

20

No

Map infinity to true and other values to false.

IsNaN

20

No

Returns which elements of the input are NaN.

LRN

13

No

Local Response Normalization proposed in the [AlexNet paper](https://papers.n

LSTM

22

No

Computes an one-layer LSTM. This operator is usually supported via some

LayerNormalization

17

No

This is layer normalization defined in ONNX as function.

LeakyRelu

16

No

LeakyRelu takes input data (Tensor ) and an argument alpha, and produces one

Less

13

No

Returns the tensor resulted from performing the Less logical operation

LessOrEqual

16

No

Returns the tensor resulted from performing the LessOrEqual logical operation

LinearAttention

27

No

Unified linear attention operator for autoregressive decoding (T=1) and prefi…

Log

13

No

Computes the Log value of the input tensor element-wise.

LogSoftmax

13

No

The operator computes the log of softmax values for the given input:

Loop

13

No

Generic Looping construct. This loop has multiple termination conditions:

LpNormalization

22

No

Given a matrix, apply Lp-normalization along the provided axis.

LpPool

22

No

LpPool consumes an input tensor X and applies Lp pooling across

MatMul

13

No

Matrix product that behaves like [numpy.matmul](https://numpy.org/doc/stable/

MatMulInteger

10

No

Matrix product that behaves like [numpy.matmul](https://numpy.org/doc/stable/

Max

13

No

Element-wise max of each of the input tensors (with Numpy-style broadcasting …

MaxPool

22

No

MaxPool consumes an input tensor X and applies max pooling across

MaxRoiPool

22

No

ROI max pool consumes an input tensor X and regions of interest (RoIs) to

MaxUnpool

22

No

MaxUnpool essentially computes the partial inverse of the MaxPool op.

Mean

13

No

Element-wise mean of each of the input tensors (with Numpy-style broadcasting…

MeanVarianceNormalization

13

No

A MeanVarianceNormalization Function: Perform mean variance normalization

MelWeightMatrix

17

No

Generate a MelWeightMatrix that can be used to re-weight a Tensor containing …

Min

13

No

Element-wise min of each of the input tensors (with Numpy-style broadcasting …

Mish

22

No

Mish: A Self Regularized Non-Monotonic Neural Activation Function.

Mod

13

No

Performs an element-wise binary modulo operation.

Mul

14

No

Performs element-wise binary multiplication (with Numpy-style broadcasting su…

Multinomial

22

No

Generate a tensor of samples from a multinomial distribution according to the…

Neg

13

No

Neg takes one input data (Tensor ) and produces one output data

NegativeLogLikelihoodLoss

22

No

A NegativeLogLikelihoodLoss operator computes (weighted) negative log likelih…

NonMaxSuppression

11

No

Filter out boxes that have high intersection-over-union (IOU) overlap with pr…

NonZero

13

No

Returns the indices of the elements that are non-zero

Not

1

No

Returns the negation of the input tensor element-wise.

OneHot

11

No

Produces a one-hot tensor based on inputs.

Optional

15

No

Constructs an optional-type value containing either an empty optional of a ce…

OptionalGetElement

18

No

If the input is a tensor or sequence type, it returns the input.

OptionalHasElement

18

No

Returns true if (1) the input is an optional-type and contains an element,

Or

7

No

Returns the tensor resulted from performing the Or logical operation

PRelu

16

No

PRelu takes input data (Tensor ) and slope tensor as input, and produces one

Pad

25

No

Given a tensor containing the data to be padded (data), a tensor containing t…

Pow

7

No

Performs element-wise exponentiation.

QLinearConv

10

No

The convolution operator consumes a quantized input tensor, its scale and zer…

QLinearMatMul

21

No

Matrix product that behaves like [numpy.matmul](https://numpy.org/doc/stable/

QuantizeLinear

25

No

The linear quantization operator consumes a high-precision tensor, a scale, a…

RMSNormalization

23

No

This is RMS normalization defined in ONNX as function as described in the pap…

RNN

22

No

Computes an one-layer simple RNN. This operator is usually supported

RandomNormal

22

No

Generate a tensor with random values drawn from a normal distribution. The shape

RandomNormalLike

22

No

Generate a tensor with random values drawn from a normal distribution.

RandomUniform

22

No

Generate a tensor with random values drawn from a uniform distribution. The s…

RandomUniformLike

22

No

Generate a tensor with random values drawn from a uniform distribution.

Range

27

No

Generate a tensor containing a sequence of numbers that begin at start and ex…

Reciprocal

13

No

Reciprocal takes one input data (Tensor ) and produces one output data

ReduceL1

18

No

Computes the L1 norm of the input tensor’s elements along the provided axes.

ReduceL2

18

No

Computes the L2 norm of the input tensor’s elements along the provided axes.

ReduceLogSum

18

No

Computes the log sum of the input tensor’s elements along the provided axes.

ReduceLogSumExp

18

No

Computes the log sum exponent of the input tensor’s elements along the provid…

ReduceMax

20

No

Computes the max of the input tensor’s elements along the provided axes.

ReduceMean

18

No

Computes the mean of the input tensor’s elements along the provided axes.

ReduceMin

20

No

Computes the min of the input tensor’s elements along the provided axes.

ReduceProd

18

No

Computes the product of the input tensor’s elements along the provided axes.

ReduceSum

13

No

Computes the sum of the input tensor’s elements along the provided axes.

ReduceSumSquare

18

No

Computes the sum square of the input tensor’s elements along the provided axes.

RegexFullMatch

20

No

RegexFullMatch performs a full regex match on each element of the input tenso…

Relu

14

No

Relu takes one input data (Tensor ) and produces one output data

Reshape

25

No

Reshape the input tensor similar to numpy.reshape.

Resize

19

No

Resize the input tensor. In general, it calculates every value in the output …

ReverseSequence

10

No

Reverse batch of sequences having different lengths specified by sequence_lens.

RoiAlign

22

No

Region of Interest (RoI) align operation described in the

RotaryEmbedding

23

No

RotaryEmbedding is the implementation of rotary positional embeddings (RoPE) …

Round

22

No

Round takes one input Tensor and rounds the values, element-wise, meaning

STFT

17

No

Computes the Short-time Fourier Transform of the signal.

Scan

11

No

Scan can be used to iterate over one or more scan_input tensors,

Scatter

11

Yes

This operator is deprecated. Please use ScatterElements, which provides the s…

ScatterElements

18

No

ScatterElements takes three inputs data, updates, and indices of the same

ScatterND

18

No

ScatterND takes three inputs data tensor of rank r >= 1, indices tensor of ra…

Selu

22

No

Selu takes one input data (Tensor ) and produces one output data

SequenceAt

11

No

Outputs a tensor copy from the tensor at ‘position’ in ‘input_sequence’.

SequenceConstruct

11

No

Construct a tensor sequence containing ‘inputs’ tensors.

SequenceEmpty

11

No

Construct an empty tensor sequence, with given data type.

SequenceErase

11

No

Outputs a tensor sequence that removes the tensor at ‘position’ from ‘input_s…

SequenceInsert

11

No

Outputs a tensor sequence that inserts ‘tensor’ into ‘input_sequence’ at ‘pos…

SequenceLength

11

No

Produces a scalar(tensor of empty shape) containing the number of tensors in …

SequenceMap

17

No

Applies a sub-graph to each sample in the input sequence(s).

Shape

25

No

Takes a tensor as input and outputs an 1D int64 tensor containing the shape o…

Shrink

9

No

Shrink takes one input data (Tensor ) and produces one Tensor output,

Sigmoid

13

No

Sigmoid takes one input data (Tensor ) and produces one output data

Sign

13

No

Calculate the sign of the given input tensor element-wise.

Sin

22

No

Computes the Sin value of the input tensor element-wise.

Sinh

22

No

Computes the Sinh value of the input tensor element-wise.

Size

25

No

Takes a tensor as input and outputs a int64 scalar that equals to the total n…

Slice

13

No

Produces a slice of the input tensor along multiple axes.

Softmax

13

No

The operator computes the normalized exponential values for the given input.

SoftmaxCrossEntropyLoss

13

No

Loss function that measures the softmax cross entropy

Softplus

22

No

Softplus takes one input data (Tensor ) and produces one output data

Softsign

22

No

Calculates the softsign (x/(1+|x|)) of the given input tensor element-wise.

SpaceToDepth

13

No

SpaceToDepth rearranges blocks of spatial data into depth. More specifically,

Split

18

No

Split a tensor into a list of tensors, along the specified ‘axis’.

SplitToSequence

11

No

Split a tensor into a sequence of tensors, along the specified ‘axis’.

Sqrt

13

No

Square root takes one input data (Tensor ) and produces one output data

Squeeze

25

No

Remove single-dimensional entries from the shape of a tensor.

StringConcat

20

No

StringConcat concatenates string tensors elementwise (with NumPy-style broadc…

StringNormalizer

10

No

StringNormalization performs string operations for basic cleaning.

StringSplit

20

No

StringSplit splits a string tensor’s elements into substrings based on a deli…

Sub

14

No

Performs element-wise binary subtraction (with Numpy-style broadcasting suppo…

Sum

13

No

Element-wise sum of each of the input tensors (with Numpy-style broadcasting …

Swish

24

No

Swish function takes one input data (Tensor ) and produces one output data (T…

Tan

22

No

Computes the Tan value of the input tensor element-wise.

Tanh

13

No

Calculates the hyperbolic tangent of the given input tensor element-wise.

TensorScatter

24

No

TensorScatter is a generic tensor update operation, motivated by the requirem…

TfIdfVectorizer

9

No

This transform extracts n-grams from the input sequence and save them as a ve…

ThresholdedRelu

22

No

ThresholdedRelu takes one input data (Tensor ) and produces one output data

Tile

13

No

Constructs a tensor by tiling a given tensor.

TopK

11

No

Retrieve the top-K largest or smallest elements along a specified axis. Given…

Transpose

25

No

Returns a transpose of the input tensor. (Similar to numpy.transpose).

Trilu

14

No

Given a 2-D matrix or batches of 2-D matrices, returns the upper or lower tri…

Unique

11

No

Find the unique elements of a tensor. When an optional attribute ‘axis’ is pr…

Unsqueeze

25

No

Insert single-dimensional entries to the shape of an input tensor (data).

Upsample

10

Yes

Upsample the input tensor.

Where

16

No

Return elements, either from X or Y, depending on condition.

Xor

7

No

Returns the tensor resulted from performing the Xor logical operation