Differences with torch.nn.functional.leaky_relu

torch.nn.functional.leaky_relu

torch.nn.functional.leaky_relu(input, negative_slope=0.01, inplace=False) -> Tensor

For more information, see torch.nn.functional.leaky_relu.

mindspore.ops.leaky_relu

mindspore.ops.leaky_relu(input, alpha=0.2) -> Tensor

For more information, see mindspore.ops.leaky_relu.

Differences

PyTorch: The leaky_relu activation function. Elements that are less than 0 in input are multiplied by negative_slope.

MindSpore: MindSpore API basically implements the same function as PyTorch. The difference is that the initial value of alpha in MindSpore is 0.2, while the corresponding negative_slope in PyTorch has an initial value of 0.01.

Categories	Subcategories	PyTorch	MindSpore	Difference
Parameters	Parameter 1	input	input	Consistent
	Parameter 2	negative_slope	alpha	Same function, different parameter names
	Parameter 3	inplace	-	Whether to make in-place changes to parameters. MindSpore does not have this feature

Code Example

# PyTorch
import torch

input = torch.tensor([-2, -1, 0, 1, 2], dtype=torch.float32)
output = torch.nn.functional.leaky_relu(input, negative_slope=0.5, inplace=False)
print(output)
# tensor([-1.0000, -0.5000,  0.0000,  1.0000,  2.0000])

# MindSpore
import mindspore

input = mindspore.Tensor([-2, -1, 0, 1, 2], dtype=mindspore.float32)
output = mindspore.ops.leaky_relu(input, alpha=0.5)
print(output)
# [-1.  -0.5  0.   1.   2. ]