Function Differences with torch.nn.GEL

View Source On Gitee

torch.nn.GELU

class torch.nn.GELU()(input)

For more information, see torch.nn.GELU.

mindspore.nn.FastGelu

class mindspore.nn.FastGelu()(input_data)

For more information, see mindspore.nn.FastGelu.

Differences

PyTorch: Cumulative distribution function based on Gaussian distribution.

MindSpore:Compared with PyTorch, MindSpore adopts a different calculation formula and has better performance.

Code Example

import mindspore
from mindspore import Tensor, nn
import torch
import numpy as np

def test_me():
    input_x = Tensor(np.array([[-1.0, 4.0, -8.0], [2.0, -5.0, 9.0]]), mindspore.float32)
    fast_gelu = nn.FastGelu()
    output = fast_gelu(input_x)
    print(output)

def test_torch():
    input_x = torch.Tensor(np.array([[-1.0, 4.0, -8.0], [2.0, -5.0, 9.0]]))
    gelu = torch.nn.GELU()
    output = gelu(input_x)
    print(output)

if __name__ == '__main__':
    test_me()
    test_torch()

# Out:
# [[-1.5419e-01  3.9922e+00 -9.7474e-06]
#  [ 1.9375e+00 -1.0053e-03  8.9824e+00]]
# tensor([[-1.5866e-01,  3.9999e+00, -0.0000e+00],
#         [ 1.9545e+00, -1.4901e-06,  9.0000e+00]])