066ef66950b6729a1cde321364b368d2fc3a9e1b,texar/modules/decoders/transformer_decoders.py,TransformerDecoder,_build,#TransformerDecoder#Any#Any#Any#,104

Before Change

                preds: [batch_size, target_length]
        
        logits = None
        if targets is not None:
            decoder_self_attention_bias = (
                attentions.attention_bias_lower_triangle(
                    layers.shape_list(targets)[1]))
            decoder_self_attention_bias = tf.Print(decoder_self_attention_bias,
                [tf.shape(decoder_self_attention_bias), decoder_self_attention_bias],
                message="decoder self attention bias",
                summarize=2048)
            target_inputs = tf.nn.embedding_lookup(self._embedding, targets)
            if self._hparams.multiply_embedding_mode == "sqrt_depth":
                target_inputs = target_inputs * \
                    (self._embedding.shape.as_list()[-1]**0.5)
            logits = self.decode(
                target_inputs,
                encoder_output,
                encoder_decoder_attention_bias,
                decoder_self_attention_bias,
            )
        preds = tf.to_int32(tf.argmax(logits, axis=-1))
        return logits, preds

    def dynamic_decode(self, encoder_output, encoder_decoder_attention_bias):

After Change

            cache=None,
        )

        logits = self.output_layer(decoder_output)
        preds = tf.to_int32(tf.argmax(logits, axis=-1))
        return logits, preds

    def dynamic_decode(self, encoder_output, encoder_decoder_attention_bias):

Instances