Linear Layers

Fully-connected linear transformation layers with various specializations. Includes standard dense layers, weight-standardized variants for improved training stability, sparse connections for efficiency, and low-rank adaptation (LoRA) for parameter-efficient fine-tuning.

Linear	Linear transformation layer.
ScaledWSLinear	Linear layer with weight standardization.
SignedWLinear	Linear layer with signed absolute weights.
SparseLinear	Linear layer with sparse weight matrix.
LoRA	Low-Rank Adaptation (LoRA) layer.
AllToAll	All-to-all connection layer.
OneToOne	One-to-one connection layer.