知识卡片
多头提供并行投影子空间
多个头用各自投影计算注意力,再拼接并输出投影,允许不同匹配模式并行表示。固定总宽度时增多头也会缩小每头维度,不能简单说头越多就必然更强或每个头都有清楚的人类语义。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 架构与学习目标
原书出处 · 1 处
- · 查看原书依据
multi-headed attention. The idea behind using multiple heads is that the model gets sufficient capacity to model various aspects of the input. The more the number of heads, the more chances that the right aspects of the input are being represented.
参考来源
- 位置:《Designing Large Language Model Applications》第04章《架构与学习目标》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch04.html)
- 结论依据:摘录中关于多头提供并行投影子空间的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。
- 原始内容:multi-headed attention. The idea behind using multiple heads is that the model gets sufficient capacity to model various aspects of the input. The more the number of heads, the more chances that the right aspects of the input are being represented.