Keras BLEU metric results in an error during LSTM training

Question

Currently, I am trying to create a simple LSTM to generate new text data. I want to use BLEU metric for the model evaluation. Both BLEU metric and LSTM are from keras library. I want to use a BLEU metric during training. However, when I put the metric in the model.compile() method, later, the model.fit() method throws an error.

I read somewhere that the error has something to do with a function not being a tensor, but I don't know how true it is and don't know what function should be converted to a tensor.

Here is the LSTM implementation with BLEU metric:

bleu = Bleu(tokenizer=tokenizer, name='bleu')

model = Sequential([
    Embedding(dictionary_length, max_len, input_length=max_len-1),
    LSTM(150),  # LSTM vrstva
    Dense(dictionary_length, activation='softmax') 
])
model.compile(loss='sparse_categorical_crossentropy', optimizer='adam', metrics=[bleu])

early_stop_callback = EarlyStopping(monitor='val_loss', patience=5)  # callback na zastavenie treningu
checkpoint_callback = ModelCheckpoint(filepath='./drive/MyDrive/colab/3best_model_v2_uni_bleu.h5', monitor='val_loss', save_best_only=True)  # callback na ulozenie najlepsieho modelu

history = model.fit(train_sequences, last_words, epochs=20, validation_split=0.2, batch_size=128, callbacks=[early_stop_callback, checkpoint_callback])

Here is the exact error that gets thrown when calling the method:

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-20-1728b7eae8ff> in <cell line: 4>()
      2 checkpoint_callback = ModelCheckpoint(filepath='./drive/MyDrive/colab/3best_model_v2_uni_bleu.h5', monitor='val_loss', save_best_only=True)  # callback na ulozenie najlepsieho modelu
      3 
----> 4 history = model.fit(train_sequences, last_words, epochs=20, validation_split=0.2, batch_size=128, callbacks=[early_stop_callback, checkpoint_callback])

7 frames
/usr/local/lib/python3.9/dist-packages/keras_nlp/metrics/bleu.py in if_body_1()
     84 
     85                                     def if_body_1():
---> 86                                         raise ag__.converted_call(ag__.ld(ValueError), (f'{ag__.ld(tensor_name)} is of rank {ag__.ld(input).shape.rank}. The last dimension must be of size 1.',), None, fscope_1)
     87 
     88                                     def else_body_1():

AttributeError: in user code:

    File "/usr/local/lib/python3.9/dist-packages/keras/engine/training.py", line 1284, in train_function  *
        return step_function(self, iterator)
    File "/usr/local/lib/python3.9/dist-packages/keras_nlp/metrics/bleu.py", line 328, in validate_and_fix_rank  *
        raise ValueError(

    AttributeError: 'function' object has no attribute 'shape'