Tag
The paper introduces Momentum DeltaNet (MDN), a linear attention model that uses stepwise momentum and parallel algorithms to improve training efficiency and performance over models like Mamba2.