pypto.normal

Applicable Products

Ascend 950PR/Ascend 950DT: Supported
Atlas A3 training products/Atlas A3 inference products: Not supported
Atlas A2 training products/Atlas A2 inference products: Not Supported

Description

Generates random numbers following a normal (Gaussian) distribution with the specified shape, where each element follows a distribution with a mean of 0 and a variance of 1.

x_i \sim N(0, 1)

Prototype

normal(shape: List[int], key: List[int], counter: List[int], alg: List[int], dtype: DataType) -> Tensor

Parameters

Parameter	Input/Output	Description
shape	Input	Shape of the output Tensor. The length supports 1 to 4 dimensions.
key	Input	Seed of the random number generator. The length only supports 1.
counter	Input	Counter of the random number generator. The length only supports 2.
alg	Input	Random number generation algorithm. Currently only value 1 (Philox algorithm) and 3 (auto_select, which selects the Philox algorithm) are supported. The length only supports 1.
dtype	Input	Data type of the output tensor. Supported data types: DT_FP32, DT_FP16, and DT_BF16.

Constraints

Splitting shape into multiple view shapes is not supported. The view shape must be consistent with the input shape.
Splitting shape into multiple tile shapes is not supported. The tile shape must be consistent with the input shape.
The last axis of the tile shape must be a multiple of 4.
counter[0] is hardcoded to 0 internally. Although the API accepts a counter list of length 2, the value of counter[0] is ignored, and the Philox counter actually used is [0, counter[1]].

Return Value

Returns a tensor with the specified shape and data type dtype, whose elements follow a normal distribution with a mean of 0 and a variance of 1.

Examples

TileShape Setting Example

Before calling this operation API, set the TileShape through set_vec_tile_shapes.

The TileShape must have the same dimension count and shape as the input.

For example, if the input shape is [m, n] and the output shape is [m, n], then TileShape should be set to [m, n].

pypto.set_vec_tile_shapes(4, 4)

API Call Example

shape = [4, 4]
key = [1234]
counter = [0, 1]
alg = [1]
dtype = pypto.DT_FP32

y = pypto.normal(shape, key, counter, alg, dtype)

The results are as follows:

Output data y: [[-0.32364845  1.8577391   0.39556974  0.2311697 ]
            [ 0.24243996 -1.9485782  -0.12983137  2.7137496 ]
            [ 1.6558666   2.0938187  -0.90338254  0.8765667 ]
            [ 0.86518306  0.01034508  0.2893259   0.01748212]]