BatchNorm2d creates a `BatchNorm2dModule` instance.

Conv2d creates a `Conv2dModule` instance TODO(qijun): only support zero padding mode only support symmetry kernel/stride/padding/dilation.

ConvTranspose2d torch.nn.conv_transpose2d TODO(qijun): only support zero padding mode only support symmetry kernel/stride/padding/dilation not support output_size when forwarding.

Functional returns a new `Functional` instance.

Linear creates a `Linear` instance.

Sequential returns a new `SequentialModule` instance.

BatchNorm2dModule torch.nn.BatchNorm2d.

Conv2dModule applies convolution over a 2D input.

ConvTranspose2dModule corresponds to torch.nn.ConvTranspose2d.

FunctionalModule wraps a function in a `Module`.

LinearModule applies a linear transformation with optional bias.

Module contains default implementation of `Module`s.

SequentialModule is a list of `Module`s that acts as a `Module` itself.

IModule is the interface of `Module`s.

Visitor is a function type supposed to be called by visitTensors.