OnnxTypes.h File Reference

Interfaces and types that are used for ONNX export. More...

#include <nvneural/CoreTypes.h>

Go to the source code of this file.

Classes
class	nvneural::IOnnxGenerationActivation
	ONNX export interface for IActivationFunction. More...
class	nvneural::IOnnxGenerationGraphAttribute
	Tool-provided interface representing an AttributeProto in an ONNX graph node. More...
class	nvneural::IOnnxGenerationGraphNode
	Tool-provided interface representing a NodeProto in an ONNX graph. More...
class	nvneural::IOnnxGenerationHost
	Tool-provided interface for creating ONNX model graphs in an ABI-stable fashion. More...
class	nvneural::IOnnxGenerationLayer
	ONNX export interface for ILayer. More...

Enumerations
enum class	nvneural::OnnxTensorDataType : std::int32_t {   Undefined = 0 , Float = 1 , Uint8 = 2 , Int8 = 3 ,   Uint16 = 4 , Int16 = 5 , Int32 = 6 , Int64 = 7 ,   String = 8 , Bool = 9 , Float16 = 10 , Double = 11 ,   Uint32 = 12 , Uint64 = 13 , Complex64 = 14 , Complex128 = 15 ,   Bfloat16 = 16 }
	Enum class wrapping ONNX TensorProto.DataType. More...

Detailed Description

Interfaces and types that are used for ONNX export.

Enumeration Type Documentation

OnnxTensorDataType

enum nvneural::OnnxTensorDataType : std::int32_t

strong

Enum class wrapping ONNX TensorProto.DataType.

Note that this is NOT nvneural::TensorDataType.

Enumerator
Float	32-bit floating point (float, TensorDataType::Float)
Uint8	8-bit unsigned integer (uint8_t)
Int8	8-bit signed integer (int8_t)
Uint16	16-bit unsigned integer (uint16_t)
Int16	16-bit signed integer (int16_t)
Int32	32-bit signed integer (int32_t, int)
Int64	64-bit signed integer (int64_t, long long)
String	Arbitrary byte array (const char*+size_t)
Bool	Boolean (bool)
Float16	16-bit floating point (__half, TensorDataType::Half)
Double	64-bit floating point (double)
Uint32	32-bit unsigned integer (uint32_t, unsigned int)
Uint64	64-bit unsigned integer (uint64_t, unsigned long long, size_t)
Complex64	Pair of 32-bit floating point values representing real/imaginary.
Complex128	Pair of 64-bit floating point values representing real/imaginary.
Bfloat16	Non-IEEE 16-bit floating-point (1 sign, 8 exponent, 7 mantissa)