问有小伙伴帮我分析一下吗？为什么相同的数据，pytorch的结果和megengine差别很大呢？？

import numpy as np

np.random.seed(2021)

x = np.random.randn(5, 3, 4, 4)

xx = np.random.randn(5, 3, 3, 1, 1)

# megengine

import megengine.functional as F

import megengine.module as M

from megengine import tenso

x_mm = tensor(x)

x_ww = tensor(xx)

x_mm = x_mm.reshape(1, -1, x_mm.shape[2], x_mm.shape[3])

x_ww = x_ww.reshape(5, 3, 3, 1, 1)

y_mm = F.conv2d(x_mm, weight=x_ww, stride=1, padding=0, groups=x_ww.shape[0])

print(y_mm.shape)

mm = y_mm.reshape(5, 3, 4, 4)

# pytorch

import torch

import torch.nn.functional as F

x_mp = torch.from_numpy(x)

x_wp = torch.from_numpy(xx)

x_mp = x_mp.reshape(1, -1, x_mp.shape[2], x_mp.shape[3])

x_wp = x_wp.reshape(5*3, 3, 1, 1)

y_mp = F.conv2d(x_mp, weight=x_wp, stride=1, padding=0, groups=5)

print(y_mp.shape)

pp = y_mp.reshape(5, 3, 4, 4)

print(mm,pp)