#Native Sparse Attention

1 stories · 0 in 7 days · 0 in 30 days

Stories, products and papers tagged "Native Sparse Attention", newest first.

Jul 13

PapersSelected09:12

COBS在NSA基础上加了个二阶统计量，长上下文检索分数从0.3拉到0.82，接近稠密注意力但省了十几倍带宽，很实用。

arXiv: DeepSeek@Alexander Tian et al. (5)Source