Definition

State Space Model (Mamba)

A state space model such as Mamba processes a sequence as a signal flowing through a dynamical system rather than comparing every token to every other. Self-attention cost grows quadratically with length; SSMs scale linearly, so million-token inputs do not explode memory. Mamba's selectivity makes its gates input-dependent, retaining important tokens and discarding noise. RWKV reaches linear time by a different route, and hybrids like Jamba mix both with Transformers.