enum GVA::Tensor::Precision

Overview

Describes tensor precision. More…

#include <tensor.h>

enum Precision {
    UNSPECIFIED = GVA_PRECISION_UNSPECIFIED,
    FP32        = GVA_PRECISION_FP32,
    FP16        = GVA_PRECISION_FP16,
    BF16        = GVA_PRECISION_BF16,
    FP64        = GVA_PRECISION_FP64,
    Q78         = GVA_PRECISION_Q78,
    I16         = GVA_PRECISION_I16,
    U4          = GVA_PRECISION_U4,
    U8          = GVA_PRECISION_U8,
    I4          = GVA_PRECISION_I4,
    I8          = GVA_PRECISION_I8,
    U16         = GVA_PRECISION_U16,
    I32         = GVA_PRECISION_I32,
    U32         = GVA_PRECISION_U32,
    I64         = GVA_PRECISION_I64,
    U64         = GVA_PRECISION_U64,
    BIN         = GVA_PRECISION_BIN,
    BOOL        = GVA_PRECISION_BOOL,
    CUSTOM      = GVA_PRECISION_CUSTOM,
};

Detailed Documentation

Describes tensor precision.

Enum Values

UNSPECIFIED

default value

FP32

32bit floating point value

FP16

16bit floating point value, 5 bit for exponent, 10 bit for mantisa

BF16

16bit floating point value, 8 bit for exponent, 7 bit for mantisa

FP64

64bit floating point value

Q78

16bit specific signed fixed point precision

I16

16bit signed integer value

U4

4bit unsigned integer value

U8

unsigned 8bit integer value

I4

4bit signed integer value

I8

8bit signed integer value

U16

16bit unsigned integer value

I32

32bit signed integer value

U32

32bit unsigned integer value

I64

64bit signed integer value

U64

64bit unsigned integer value

BIN

1bit integer value

BOOL

8bit bool type

CUSTOM

custom precision has it’s own name and size of elements