fdc2749cceb52addb817f22ed5531535867cd29e,examples/attention_tagger.py,,main,#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#,138

Before Change

                >> Maxout(width, pieces=3), pad=0)
            >> PositionEncode(1000, width)
            >> flatten_add_lengths
            >> MultiHeadedAttention(nM=width, nH=8)
            >> with_getitem(0, LayerNorm(nO=width))
            >> unflatten
            >> with_flatten(Softmax(nr_tag))
        )

After Change

                >> Maxout(width, pieces=3), pad=0)
            >> PositionEncode(1000, width)
            >> flatten_add_lengths
            >> Residual(MultiHeadedAttention(nM=width, nH=1))
            >> unflatten
            >> with_flatten(Softmax(nr_tag))
        )

Instances