enum GVAPrecision

Overview

This enum describes model layer precision. More…

#include <gva_tensor_meta.h>

enum GVAPrecision {
    GVA_PRECISION_UNSPECIFIED = 255,
    GVA_PRECISION_FP32        = 10,
    GVA_PRECISION_FP16        = 11,
    GVA_PRECISION_BF16        = 12,
    GVA_PRECISION_FP64        = 13,
    GVA_PRECISION_Q78         = 20,
    GVA_PRECISION_I16         = 30,
    GVA_PRECISION_U4          = 39,
    GVA_PRECISION_U8          = 40,
    GVA_PRECISION_I4          = 49,
    GVA_PRECISION_I8          = 50,
    GVA_PRECISION_U16         = 60,
    GVA_PRECISION_I32         = 70,
    GVA_PRECISION_U32         = 74,
    GVA_PRECISION_I64         = 72,
    GVA_PRECISION_U64         = 73,
    GVA_PRECISION_BIN         = 71,
    GVA_PRECISION_BOOL        = 41,
    GVA_PRECISION_CUSTOM      = 80,
};

Detailed Documentation

This enum describes model layer precision.

Enum Values

GVA_PRECISION_UNSPECIFIED

default value

GVA_PRECISION_FP32

32bit floating point value

GVA_PRECISION_FP16

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

GVA_PRECISION_BF16

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

GVA_PRECISION_FP64

64bit floating point value

GVA_PRECISION_Q78

16bit specific signed fixed point precision

GVA_PRECISION_I16

16bit signed integer value

GVA_PRECISION_U4

4bit unsigned integer value

GVA_PRECISION_U8

unsignned 8bit integer value

GVA_PRECISION_I4

4bit signed integer value

GVA_PRECISION_I8

8bit signed integer value

GVA_PRECISION_U16

16bit unsigned integer value

GVA_PRECISION_I32

32bit signed integer value

GVA_PRECISION_U32

32bit unsigned integer value

GVA_PRECISION_I64

64bit signed integer value

GVA_PRECISION_U64

64bit unsigned integer value

GVA_PRECISION_BIN

1bit integer value

GVA_PRECISION_BOOL

8bit bool type

GVA_PRECISION_CUSTOM

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