Как исправить «ValueError: не удалось передать входной массив (2, ...) в (1, ...)» model.predict?

Question

После обучения сети Encoder Decoder я использую ее для прогнозирования. Однако, предоставляя те же данные, которые я использовал для обучения, я получаю ошибку значения.

Я пробовал установить 'batch_size = 1'. Я пытался запустить его только на одном входе.

def build(embedding_size, hidden_size):
    encoder_inputs = Input(shape=(None, embedding_size))
    encoder = Bidirectional(LSTM(hidden_size, return_state=True))
    encoder_outputs, forward_h, forward_c, backward_h, backward_c = encoder(encoder_inputs)
    encoder_states = [tf.concat([forward_h, backward_h], 0), tf.concat([forward_c, backward_h], 0)]

    decoder_inputs = Input(shape=(None, embedding_size))
    decoder_lstm = LSTM(hidden_size, return_sequences=True, return_state=True)
    decoder_outputs, _, _ = decoder_lstm(decoder_inputs,
                                         initial_state=encoder_states)
    decoder_dense = Dense(embedding_size, activation='relu')
    decoder_outputs = decoder_dense(decoder_outputs)

    model = Model([encoder_inputs, decoder_inputs], decoder_outputs)

    return model


model = build(**params)
model.load_weights('model/model-01.hdf5')

#Shape of encoder_input_data is (2752, 43, 241)
#Shape of decoder_input_data is (2752, 43, 241)

predictions = model.predict([encoder_input_data, decoder_input_data], batch_size=1)

Traceback (most recent call last):
  File "C:/Users/mason-act5/NLP/mrs_paraphrase/predict.py", line 48, in <module>
    predictions = model.predict([encoder_input_data, decoder_input_data], batch_size=1)
  File "C:\Users\mason-act5\NLP\rnn\venv\lib\site-packages\tensorflow\python\keras\engine\training.py", line 1078, in predict
    callbacks=callbacks)
  File "C:\Users\mason-act5\NLP\rnn\venv\lib\site-packages\tensorflow\python\keras\engine\training_arrays.py", line 370, in model_iteration
    aggregator.aggregate(batch_outs, batch_start, batch_end)
  File "C:\Users\mason-act5\NLP\rnn\venv\lib\site-packages\tensorflow\python\keras\engine\training_utils.py", line 169, in aggregate
    self.results[i][batch_start:batch_end] = batch_out
ValueError: could not broadcast input array from shape (2,43,241) into shape (1,43,241)