This is an implementation of multi-headed attention based on "Attention is all you Need". If query, key, value are the same, then this is self-attention. Each timestep in query attends to the corresponding sequence in key, and returns a fixed-width vector.
layer_multi_head_attention(
inputs,
num_heads,
key_dim,
value_dim = NULL,
dropout = 0,
use_bias = TRUE,
output_shape = NULL,
attention_axes = NULL,
kernel_initializer = "glorot_uniform",
bias_initializer = "zeros",
kernel_regularizer = NULL,
bias_regularizer = NULL,
activity_regularizer = NULL,
kernel_constraint = NULL,
bias_constraint = NULL,
...
)attention_output: The result of the computation, of shape [B, T, E], where
T is for target sequence shapes and E is the query input last dimension if
output_shape is None. Otherwise, the multi-head outputs are project to the
shape specified by output_shape.
attention_scores: (Optional) multi-head attention coeffients over attention axes.
query: Query Tensor of shape [B, T, dim].
value: Value Tensor of shape [B, S, dim].
key: Optional key Tensor of shape [B, S, dim]. If not given, will use value
for both key and value, which is the most common case.
attention_mask: a boolean mask of shape [B, T, S], that prevents attention
to certain positions.
return_attention_scores: A boolean to indicate whether the output should be attention output if TRUE, or (attention_output, attention_scores) if FALSE. Defaults to FALSE.
training: Python boolean indicating whether the layer should behave in training mode (adding dropout) or in inference mode (no dropout). Defaults to either using the training mode of the parent layer/model, or FALSE (inference) if there is no parent layer.
This layer first projects query, key and value. These are (effectively) a list
of tensors of length num_attention_heads, where the corresponding shapes are
[batch_size, , key_dim], [batch_size, , key_dim], [batch_size, , value_dim].
Then, the query and key tensors are dot-producted and scaled. These are softmaxed to obtain attention probabilities. The value tensors are then interpolated by these probabilities, then concatenated back to a single tensor.
Finally, the result tensor with the last dimension as value_dim can take an linear projection and return.