4ce9a9d078d41af1a960f0e5bf16e373f69d5117,open_seq2seq/decoders/rnn_decoders.py,RNNDecoderWithAttention,_decode,#RNNDecoderWithAttention#Any#,151

Before Change

      self._tgt_vocab_size, use_bias=False,
    )

    cell_params = copy.deepcopy(self.params)
    cell_params["num_units"] = self.params["decoder_cell_units"]

    if self._mode == "train":
      dp_input_keep_prob = self.params["decoder_dp_input_keep_prob"]
      dp_output_keep_prob = self.params["decoder_dp_output_keep_prob"]
    else:
      dp_input_keep_prob = 1.0
      dp_output_keep_prob = 1.0

    if self.params["attention_type"].startswith("gnmt"):
      residual_connections = False
      wrap_to_multi_rnn = False
    else:
      residual_connections = self.params["decoder_use_skip_connections"]
      wrap_to_multi_rnn = True

    self._decoder_cells = create_rnn_cell(
      cell_type=self.params["decoder_cell_type"],
      cell_params=cell_params,
      num_layers=self.params["decoder_layers"],
      dp_input_keep_prob=dp_input_keep_prob,
      dp_output_keep_prob=dp_output_keep_prob,
      residual_connections=residual_connections,
      wrap_to_multi_rnn=wrap_to_multi_rnn,
    )

    attention_mechanism = self._build_attention(
      encoder_outputs,
      enc_src_lengths,

After Change

    // list of cells
    self._decoder_cells = [
      single_cell(cell_class=self.params["core_cell"],
                  cell_params=self.params.get("core_cell_params", {}),
                  dp_input_keep_prob=dp_input_keep_prob,
                  dp_output_keep_prob=dp_output_keep_prob,
                  // residual connections are added a little differently for GNMT
                  residual_connections=False if self.params["attention_type"].startswith("gnmt") else residual_connections,
                  ) for _ in range(self.params["decoder_layers"])]

    attention_mechanism = self._build_attention(
      encoder_outputs,

Instances