Expressivity-Efficiency Tradeoffs for Hybrid Sequence Models

ORID 82EJxJzG6r · tags icml2026-repro paper-82EJxJzG6r

#StatusPageArtifactClaim excerpt
1VERIFIED 2/201-any-k-layer-state-space-model-solvingartifactTheorem 3.3 proves that any k-layer state-space model solving the function-compo…
2VERIFIED 2/202-sliding-window-transformers-solving-same-tasksartifactTheorem 3.7 proves that sliding-window Transformers solving the same tasks under…
3VERIFIED 2/203-constructs-two-layer-hybrid-mamba-attentionartifactTheorem 4.3 constructs a two-layer hybrid (Mamba + attention) model that solves …
4VERIFIED 2/204-constructs-three-layer-hybrid-model-accuracyartifactTheorem 4.6 constructs a three-layer hybrid model that achieves 99% accuracy on …
5VERIFIED 2/205-selective-copying-task-learned-hybridartifactOn the selective copying task, the learned hybrid model reaches perfect accuracy…
6VERIFIED 2/206-multi-key-associative-recall-hybrid-modelartifactOn multi-key associative recall, the hybrid model reaches 60% accuracy using 6x …

Open logbook index · logbook.json