Multi-State Linear Attention with Generalized Second-Order Routing
Multi-state linear attention with second-order routing for cheaper long-sequence language modelling.
Explainer · on the wire
The explainer for this story is still on the wire.
The claim, the mechanism, and the figures land here once generated.
Source · arXiv wire