71bd73748b835de5ae20bdc90ce4321e47f4c2b2,python/eight_mile/tf/layers.py,EmbeddingsStack,call,#EmbeddingsStack#Any#,390

Before Change

        :return: A 3-d vector where the last dimension is the concatenated dimensions of all embeddings
        
        all_embeddings_out = []
        for k, embedding in self.embeddings.items():
            x = inputs[k]
            embeddings_out = embedding(x)
            all_embeddings_out.append(embeddings_out)
        word_embeddings = tf.concat(values=all_embeddings_out, axis=-1)
        return self.dropout(word_embeddings, TRAIN_FLAG())

    @property

After Change

        parallels = []
        expanded = tf.expand_dims(inputs, ParallelConv.DUMMY_AXIS)
        for W, b in zip(self.Ws, self.bs):
            conv = tf.nn.conv2d(
                expanded, W,
                strides=[1, 1, 1, 1],
                padding="SAME", name="CONV"
            )
            activation = self.activation(tf.nn.bias_add(conv, b), "activation")
            parallels.append(activation)
        combine = tf.reshape(tf.concat(values=parallels, axis=ParallelConv.FEATURE_AXIS), [-1, self.output_dim])
        return combine

    def compute_output_shape(self, input_shape):

Instances