torch 2.14.0 | Python 3.12.13 | CPU
mode   no_grad  same_output  output_requires_grad  weight_requires_grad
train  False    False        True                  True
train  True     False        False                 True
eval   False    True         True                  True
eval   True     True         False                 True
eval backward weight gradient row: [6.0, 8.0, 10.0, 12.0]
BatchNorm running_mean after one no_grad forward:
train: [1.2, 2.2]
eval: [0.0, 0.0]