NVIDIA NvNeural SDK  2022.2
GPU inference framework for NVIDIA Nsight Deep Learning Designer
nvneural::ActivationDesc Struct Reference

Structure describing an activation function (IActivationFunction) for tool interfaces and network builders. More...

#include <nvneural/CoreTypes.h>

Public Attributes

ActivationFunctionId activationFunction
 This is the activation function ID implemented by this activation function.
 
NetworkBackendId backend
 This is the backend identifier supported by this instance of the activation function. More...
 
std::size_t numParameters
 This is the number of parameters supported for this activation. More...
 
const char * pDisplayName
 An optional human-readable layer name. More...
 
const char * pDocumentation
 An optional human-readable descriptive text. More...
 
const char * pObjectClass
 This is the programmatic class name used to instantiate the activation object through the IPlugin::createObject interface. More...
 
const ParamDescpParameters
 This is the pointer to the contiguous memory of ParamDesc structs. More...
 
const char * pSerializedName
 This is the name used to represent the activation function in serialized form. More...
 
std::size_t structSize
 This is the sizeof() this struct. More...
 
TensorFormat tensorFormat
 This is the tensor format supported by this activation function. More...
 

Detailed Description

Structure describing an activation function (IActivationFunction) for tool interfaces and network builders.

All contained pointers must be valid for the lifetime of the relevant IPlugin object.

Member Data Documentation

◆ backend

NetworkBackendId nvneural::ActivationDesc::backend

This is the backend identifier supported by this instance of the activation function.

Activation functions supporting multiple backends should expose multiple ActivationDesc instances.

◆ numParameters

std::size_t nvneural::ActivationDesc::numParameters

This is the number of parameters supported for this activation.

Those are described in the ParamDesc structure.

◆ pDisplayName

const char* nvneural::ActivationDesc::pDisplayName

An optional human-readable layer name.

Generally used in host editor. If this is not defined (nullptr) then pSerializedName will be used. String should be treated as UTF-8.

Example: "Leaky ReLU"

◆ pDocumentation

const char* nvneural::ActivationDesc::pDocumentation

An optional human-readable descriptive text.

Generally used in host editor. String should be treated as UTF-8.

◆ pObjectClass

const char* nvneural::ActivationDesc::pObjectClass

This is the programmatic class name used to instantiate the activation object through the IPlugin::createObject interface.

Example: "com.nvidia.activations.relu"

◆ pParameters

const ParamDesc* nvneural::ActivationDesc::pParameters

This is the pointer to the contiguous memory of ParamDesc structs.

The number of these structs in the memory block is defined in numParameters. Each struct describes a particular parameter.

◆ pSerializedName

const char* nvneural::ActivationDesc::pSerializedName

This is the name used to represent the activation function in serialized form.

Examples: "relu", "leaky_tanh", "leaky_sigmoid"

◆ structSize

std::size_t nvneural::ActivationDesc::structSize

This is the sizeof() this struct.

It is used for versioning, for instance, one plugin could have an earlier version of this struct and the host a later version and thus, their structSize would be different.

◆ tensorFormat

TensorFormat nvneural::ActivationDesc::tensorFormat

This is the tensor format supported by this activation function.

Activation functions supporting multiple tensor formats should expose multiple ActivationDesc instances.


The documentation for this struct was generated from the following file: