知识卡片
逐位置前馈不混合相邻词元
常见Transformer前馈网络对每个位置独立使用共享参数,经过升维、非线性与降维变换;跨位置交互主要由注意力等模块承担。共享参数不等于所有位置输出相同,因为输入上下文向量不同。
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
Each token representation is independently fed through the network. The feedforward network incorporates a nonlinear activation function like Rectified Linear Unit (ReLU) or Gaussian Error Linear Units (GELU), thus enabling the model to learn more complex features from the data. For more details on these activation functions, refer to this blog post from v7.
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:摘录中关于逐位置前馈不混合相邻词元的定义、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:Each token representation is independently fed through the network. The feedforward network incorporates a nonlinear activation function like Rectified Linear Unit (ReLU) or Gaussian Error Linear Units (GELU), thus enabling the model to learn more complex features from the data. For more details on these activation functions, refer to this blog post from v7.