Skip to content

dot_product_matrix_attention

allennlp.modules.matrix_attention.dot_product_matrix_attention

[SOURCE]


DotProductMatrixAttention#

@MatrixAttention.register("dot_product")
class DotProductMatrixAttention(MatrixAttention)

Computes attention between every entry in matrix_1 with every entry in matrix_2 using a dot product.

Registered as a MatrixAttention with name "dot_product".

forward#

class DotProductMatrixAttention(MatrixAttention):
 | ...
 | @overrides
 | def forward(
 |     self,
 |     matrix_1: torch.Tensor,
 |     matrix_2: torch.Tensor
 | ) -> torch.Tensor