f81e27a5cbe72ec3bf22f12d5a9ea60008bb8f1c,allennlp/modules/seq2seq_encoders/stacked_self_attention.py,StackedSelfAttentionEncoder,forward,#StackedSelfAttentionEncoder#Any#Any#,117

Before Change

                // layers, so we exclude it here.
                feedforward_output += cached_input
            // shape (batch_size, sequence_length, hidden_dim)
            attention_output = attention(layer_norm(feedforward_output), mask)
            output = self.dropout(attention_output) + feedforward_output
        return self._output_layer_norm(output)

After Change

            if feedforward_output.size() == cached_input.size():
                // First layer might have the wrong size for highway
                // layers, so we exclude it here.
                feedforward_output = feedforward_layer_norm(feedforward_output + cached_input)
            // shape (batch_size, sequence_length, hidden_dim)
            attention_output = attention(feedforward_output, mask)
            output = layer_norm(self.dropout(attention_output) + feedforward_output)

        return output

    @classmethod

Instances