1 paper
Zian Liu, Yiwen Hu, Zican Dong +5
Recently, linear attention layers have been increasingly adopted to replace softmax attention at scale for long-context modeling. However, existing context extension approaches typ…