API Reference

This section contains the API reference for the Warp-NN library.

Core modules

Buffer	Class representing a non-learnable state array (e.g. running statistics).
Module	Base abstract class for all the modules.
Parameter	Class representing a learnable parameter.

Activations

CELU	Continuously differentiable Exponential Linear Unit (CELU) activation function.
ELU	Exponential Linear Unit (ELU) activation function.
GELU	Gaussian Error Linear Unit (GELU) activation function.
HardSigmoid	Hard Sigmoid activation function.
HardSwish	Hard Swish (HardSwish) activation function.
LeakyReLU	Leaky Rectified Linear Unit (Leaky ReLU) activation function.
LogSoftmax	Log-Softmax activation function.
Mish	Mish activation function.
PReLU	Parametric ReLU (PReLU) operation.
ReLU	Rectified Linear Unit (ReLU) activation function.
SELU	Scaled Exponential Linear Unit (SELU) activation function.
Shrink	Shrink activation function.
Sigmoid	Sigmoid activation function.
Softmax	Softmax activation function.
Softplus	Softplus activation function.
Softsign	Softsign activation function.
Swish	Swish activation function.
Threshold	Threshold activation function.

Initializers

constant	Initialize the array with a constant value.
kaiming_normal	Initialize the array using the Kaiming (aka He) normal initialization method.
kaiming_uniform	Initialize the array using the Kaiming (aka He) uniform initialization method.
ones	Initialize the array with ones.
zeros	Initialize the array with zeros.

Layers

AvgPool1D	Apply a 1D average pooling.
AvgPool2D	Apply a 2D average pooling.
BatchNorm	Apply Batch Normalization over the channels of the input.
Conv1D	Apply a 1D convolution.
Conv2D	Apply a 2D convolution.
Dropout	Randomly zero some of the input values during training.
Flatten	Flatten a contiguous range of dimensions into a single dimension.
GlobalAvgPool	Apply a global average pooling over the spatial dimensions of the input.
GlobalMaxPool	Apply a global max pooling over the spatial dimensions of the input.
GroupNorm	Apply Group Normalization over the channels of the input.
GRU	Apply a multi-layer Gated Recurrent Unit (GRU) to an input sequence.
GRUCell	Apply a Gated Recurrent Unit (GRU) cell.
Identity	Identity operation.
InstanceNorm	Apply Instance Normalization over the spatial dimensions of the input.
LayerNorm	Apply Layer Normalization over the trailing dimensions of the input.
Linear	Apply a linear transformation over the final dimension of the input.
LSTM	Apply a multi-layer Long Short-Term Memory (LSTM) to an input sequence.
LSTMCell	Apply a Long Short-Term Memory (LSTM) cell.
MaxPool1D	Apply a 1D max pooling.
MaxPool2D	Apply a 2D max pooling.
RMSNorm	Apply Root Mean Square (RMS) Normalization over the trailing dimensions of the input.
RNN	Apply a multi-layer Elman's Recurrent Neural Network (RNN) to an input sequence.
RNNCell	Apply a Elman's Recurrent Neural Network (RNN) cell.
Sequential	Apply callable modules (e.g. layers, activation functions, etc.) connected in a cascading sequence.

Operators

Abs	Absolute Value (Abs) operation.
Acos	Arccosine (Acos) operation.
Acosh	Inverse Hyperbolic Cosine (Acosh) operation.
Add	Addition (Add) operation.
Asin	Arcsine (Asin) operation.
Asinh	Inverse Hyperbolic Sine (Asinh) operation.
Atan	Arctangent (Atan) operation.
Atanh	Inverse Hyperbolic Tangent (Atanh) operation.
BitwiseAnd	Bitwise AND (BitwiseAnd) operation.
BitwiseNot	Bitwise NOT (BitwiseNot) operation.
BitwiseOr	Bitwise OR (BitwiseOr) operation.
BitwiseXor	Bitwise XOR (BitwiseXor) operation.
Ceil	Ceiling (Ceil) operation.
Clip	Element-wise clipping (clamping) of the input values into the interval [min_val, max_val].
Cos	Cosine (Cos) operation.
Cosh	Hyperbolic Cosine (Cosh) operation.
Div	Division (Div) operation.
Erf	Error Function (Erf) operation.
Exp	Exponential (Exp) operation.
Floor	Floor operation.
Log	Natural Logarithm (Log) operation.
Max	Element-wise maximum (Max) operation.
Min	Element-wise minimum (Min) operation.
Mul	Multiplication (Mul) operation.
Neg	Negation (Neg) operation.
Pow	Power (Pow) operation.
Reciprocal	Reciprocal operation.
Round	Round operation.
Sign	Sign operation.
Sin	Sine (Sin) operation.
Sinh	Hyperbolic Sine (Sinh) operation.
Sqrt	Square Root (Sqrt) operation.
Sub	Subtraction (Sub) operation.
Tan	Tangent (Tan) operation.
Tanh	Hyperbolic Tangent (Tanh) operation.

Optimizers

Adam	Adam optimizer.
SGD	SGD optimizer.

Runtime

OnnxRuntime

Execute an ONNX graph using Warp-NN modules.