problem with crf layer

[jozi]

this is my model

    model = Sequential()
    model.add(Bidirectional(LSTM(CONFIG.hidden_size, return_sequences=True),input_shape=(CONFIG.max_input_len,CONFIG.number_of_chars)))
    model.add(Dropout(CONFIG.amount_of_dropout))
    model.add(TimeDistributed(Dense(n_classes, activation='softmax'),input_shape=(CONFIG.max_input_len, CONFIG.number_of_chars)))
    model.add(Dropout(0.2))
    crf = CRFLayer()
    model.add(crf(inputs=[CONFIG.max_input_len,CONFIG.number_of_chars]))
    model.compile(loss=crf.loss, optimizer='rmsprop', metrics=['accuracy'])

and get a error
ValueError: Layer crf_layer_1 was called with an input that isn't a symbolic tensor. Received type: <class 'int'>. Full input: [60, 53]. All inputs to the layer should be tensors.
below work nice without crf

    model = Sequential()
    model.add(Bidirectional(LSTM(CONFIG.hidden_size, return_sequences=True),input_shape=(CONFIG.max_input_len,CONFIG.number_of_chars)))
    model.add(Dropout(CONFIG.amount_of_dropout))
    model.add(TimeDistributed(Dense(n_classes,activation='softmax'), input_shape=(CONFIG.max_input_len, CONFIG.number_of_chars)))

    model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['categorical_accuracy'])