SMat-Attention: Structured Long-Context Sequence Modeling
arXiv:2609.36062v1 Announce Type: new Abstract: Long-context sequence models face a fundamental tradeoff: softmax attention uses flexible token-level interactions at quadratic cost, whereas linear attention obtains linear-time training and...
Read more: https://arxiv.org/abs/2609.36062
Scan with your phone
Scan with your phone