#ssm

Distilling Transformers into Mamba via Linearized Attention

A two-stage knowledge transfer method preserves Transformer performance in State Space Models by routing through linearized attention as an intermediate step.