Mamba-2
Summary
Mamba-2 is the structured state space duality architecture that reframes SSMs as semiseparable matrix mixers and uses the SSD algorithm for faster, more hardware-friendly recurrent sequence modeling.
Role In The Wiki
Mamba-2 is the bridge between attention-style matrix views and recurrent SSM views. It is the immediate linear-recurrence background for ParaRNN’s nonlinear parallel-training claim.