Infini-attention实现无限长文本处理
谷歌推出Infini-attention机制，使Transformer模型在有限计算资源下处理无限长输入，实现114倍内存压缩比，显著提升长文本处理能力并刷新SOTA记录。https://mp.weixin.qq.com/s?__biz=MzIzNjc1NzUzMw==&mid=2247725133&idx=3&sn=33f83691581506af5724f98821f5d5f7&chksm=e9647895b5d6927fa3074123609b046f5430dda2b29eaa934d110dcd6b9e4c9936496f722ced&scene=126&sessionid=1712991605#rd
页:
[1]