Module builder 

Model builder.

MlpBuilder is the recommended way to define a model.

It makes model structure explicit (layer sizes + activations) and chooses a reasonable default weight initializer for each activation:

• tanh / sigmoid / identity: Xavier/Glorot
• relu / leaky relu: He/Kaiming

The resulting Mlp still supports the low-level, allocation-free hot path: reuse Scratch / Gradients for per-sample forward/backward.

Structs

MlpBuilder

Builder for an Mlp.