Triadic Linear Attention: Three-Dimensional Recurrent States for Long-Context Sequence Modeling
Triadic linear attention extends linear attention to third-order tensor states, improving long-context modeling and recall over state-enlargement alternatives.