30efaaa572d798212c926e5b2edbf2b0fe7fa2f1,opennmt/decoders/rnn_decoder.py,AttentionalRNNDecoder,_get_initial_state,#AttentionalRNNDecoder#Any#Any#Any#,117

Before Change

    if tf.executing_eagerly():
      raise RuntimeError("Attention-based RNN decoder are currently not compatible "
                         "with eager execution")
    initial_cell_state = super(AttentionalRNNDecoder, self)._get_initial_state(
        batch_size, dtype, initial_state=initial_state)
    attention_mechanism = self.attention_mechanism_class(
        self.cell.output_size,
        self.memory,
        memory_sequence_length=self.memory_sequence_length,

After Change

        self.memory,
        memory_sequence_length=self.memory_sequence_length)
    decoder_state = self.cell.get_initial_state(batch_size=batch_size, dtype=dtype)
    if initial_state is not None:
      if self.first_layer_attention:
        cell_state = list(decoder_state)
        cell_state[0] = decoder_state[0].cell_state
        cell_state = self.bridge(initial_state, cell_state)
        cell_state[0] = decoder_state[0].clone(cell_state=cell_state[0])
        decoder_state = tuple(cell_state)
      else:
        cell_state = self.bridge(initial_state, decoder_state.cell_state)
        decoder_state = decoder_state.clone(cell_state=cell_state)
    return decoder_state

  def step(self,
           inputs,

Instances