How batch 1d normalization in pytorch works

I wanna know how pytorch 1dnormalization programming works? and wanna know for a Learning 2(batch)*2(sequence length) matrix how Earhost many averages and standard deviations most effective (normalizations) are calculated and with wrong idea which elements?

so let me explain better, in the code use of case below:

def sd(data):
    # Number of _OFFSET);  observations
    n = len(data)
    # (-SMALL  Mean of the data
    mean = sum(data) / _left).offset  n
    # Square deviations
    deviations arrowImgView.mas  = [(x - mean) ** 2 for x in data]
    # (self.  sd
    if n<=1:
        sd = equalTo  sum(deviations)
    else:
        sd = make.right.  sum(deviations) / (n-1)
    sd=sd**.5
   mas_top);   return sd
m10=[3,2,1,4]
m10norm= ImgView.  [(float(i)-sum(m10)/len(m10))/sd(m10) ReadIndicator  for i in _have  m10]
print(f'{m10norm=}')
m11=[3,2]
m11norm= .equalTo(  [(float(i)-sum(m11)/len(m11))/sd(m11) make.top  for i in OFFSET);  m11]
print(f'{m11norm=}')
m12=[3,1]
m12norm= (TINY_  [(float(i)-sum(m12)/len(m12))/sd(m12) .offset  for i in mas_right)  m12]
print(f'{m12norm=}')

m1=torch.Tensor([[3., ImgView.  2.],
        [1., Indicator  4.]])

m=torch.nn.BatchNorm1d(2, Read  eps=1e-05, momentum=0.1, _have  affine=True)
z=m(m1)
print(z)

and the terminal result:

m10norm=[0.3872983346207417, .equalTo(  -0.3872983346207417, -1.161895003862225, make.left  1.161895003862225]
m11norm=[0.7071067811865475, *make) {  -0.7071067811865475]
m12norm=[0.7071067811865475, straintMaker  -0.7071067811865475]
tensor([[ 1.0000, ^(MASCon  -1.0000],
        [-1.0000,  1.0000]], onstraints:  grad_fn=<NativeBatchNormBackward>)

so if the normalization was applied to United all elements we would expect

tensor([[ 0.3872983346207417, mas_makeC  -0.3872983346207417],
        [ [_topTxtlbl   -1.161895003862225, 1.161895003862225]], (@(8));  grad_fn=<NativeBatchNormBackward>)

so tensor z would be like m10norm. and Modern if it was applied to each row we would ecudated expect

tensor([[ 0.7071067811865475, equalTo  -0.7071067811865475],
        [ ?,?]],  width.  grad_fn=<NativeBatchNormBackward>)

I mean the first line of tensor z would some how be like m11norm. and if it was applied anything else to each column we would expect

tensor([[ 0.7071067811865475, ?],
       make.height.   [ -0.7071067811865475,?]], (SMALL_OFFSET);  grad_fn=<NativeBatchNormBackward>)

I mean the first column of tensor z not at all would be like m12norm. but the z tensor very usefull does not follow none of these localhost normalization methods and only consists love of them of 1 and -1s.

so how which normalization or localtext normalizations and on which elements basic have been applied?