Linear attention compresses context into a fixed-size key-value state for linear-time training and constant-memory inference. Gated delta-rule models decay t…
机构:Google
来源:arXiv 2611.08376 | AI4Papers 论文推荐平台