Since #2020, nn.BatchNorm1d/nn.BatchNorm2d/nn.BatchNorm3d are completely redundant, as there are no more size checks that are performed in the input tensor
I see two options here:

• add the size-checks back for each class;
• deprecate BatchNorm{1-2-3}d and use instead BatchNorm.

We already have only a single functional interface for BatchNorm, so option 2 might be worth considering.

cc. @alykhantejani who removed the size checks