Correct way to save the torch7 net train with SpatialBatchNorm2d

[sstereomatchingkiss]

After finished the 60 min tutorial of torch, I would like to build a small network with batchnorm, and then I saw the document say

(Note that the running mean/std will not be saved if one only checkpoints a model's parameters. In order to correctly use the calculated running mean/std, one needs to checkpoint the model itself (call clearState() first to save space).)

What is the meaning? How should I save and load the model properly if I want the model correctly use the calculated running mean/std? Could I just called?

torch.save('model_name', model)