来自Trax的AttentionQKV

Question

Trax实现的AttentionQKV层如下所示：AttentionQKV

def AttentionQKV(d_feature, n_heads=1, dropout=0.0, mode='train'):
  """Returns a layer that maps (q, k, v, mask) to (activations, mask).
  See `Attention` above for further context/details.
  Args:
    d_feature: Depth/dimensionality of feature embedding.
    n_heads: Number of attention heads.
    dropout: Probababilistic rate for internal dropout applied to attention
        activations (based on query-key pairs) before dotting them with values.
    mode: One of `'train'`, `'eval'`, or `'predict'`.
  """
  return cb.Serial(
      cb.Parallel(
          core.Dense(d_feature),
          core.Dense(d_feature),
          core.Dense(d_feature),
      ),
      PureAttention(  # pylint: disable=no-value-for-parameter
          n_heads=n_heads, dropout=dropout, mode=mode),
      core.Dense(d_feature),
  )

特别是，三个平行致密层的目的是什么？这个层的输入是q，k，v，掩码。为什么q，k，v被放置在稠密层中？

Answer 1

这个代码片段是注意是你所需要的文件第5页上等式的一个实现，该文件在2017年引入了变压器模型。计算结果如图2所示：

​

​

隐藏状态被投影到h注意头中，从而并行地进行缩放的点积注意。投影可以解释为提取与头部相关的信息。然后，每个头根据不同的(学习)标准进行概率检索。