MultinomialOperation

Hardware Model	Supported
Atlas 350 accelerator cards	x
Atlas A3 inference products/Atlas A3 training products	√
Atlas A2 training products/Atlas A2 inference products	√
Atlas training products	x
Atlas inference products	x
Atlas 200I/500 A2 inference products	x

Samples the last axis by randomly extracting numSamples values and outputting the index.

struct MultinomialParam {
    uint32_t numSamples = 1;
    uint32_t randSeed = 0;
    uint8_t rsv[8] = {0};
};

Member	Type	Default Value	Description
numSamples	uint32_t	1	Number of random samples. The value must be less than or equal to the dimension size of x. The maximum value is 64. The value must be greater than 0 but less than vocSize.
randSeed	uint32_t	0	Random number seed.
rsv[8]	uint8_t	{0}	Reserved

Member

Type

Default Value

Description

numSamples

uint32_t

Number of random samples.

The value must be less than or equal to the dimension size of x. The maximum value is 64.
The value must be greater than 0 but less than vocSize.

randSeed

uint32_t

Random number seed.

rsv[8]

uint8_t

{0}

Reserved

Parameter	Dimension	Data Type	Format	Description
x	[batch, vocSize]	float16	ND	Input tensor. If the input of the last dimension is greater than 98304, a precision error may occur.

Parameter	Dimension	Data Type	Format	Description
output	[batch, numSamples]	int32	ND	Index of the sampled data. dims[1] must be equal to numSamples.

Ensure that the last axis is normalized.

Input

intensor =

[[0.0701, 0.1021, 0.1010, 0.1005],
[0.1101, 0.1020, 0.1017, 0.0990],
[0.0592, 0.1006, 0.1006, 0.1002]]

numSamples = 1

randSeed = 0

Parent topic: atb/infer_op_params.h