Mamba-2

"Transformers are SSMs" (ICML 2024): Albert Gu (CMU) and Tri Dao (Princeton) unify attention and state-space models via structured state-space duality — the theoretical and practical basis of the hybrid SSM blocks now inside production models from Nemotron-H to the Kimi/Ant linear-attention lines (state-spaces repo: 18.6K★). Filed jointly with Princeton by author affiliation per our project-org rule; the original Mamba (2312.00752, Dec 2023) is the pre-window anchor of the same line, and MOHAWK (2024) distills Transformers into SSMs.