How to use ELMo in Keras version?

For some reason, I have to use ELMo in Keras platform. When I try to write a Keras Layer for ELMo, I notice that when the input of **BidirectionalLanguageModel** is keras.layer.Input, the whole program will be stuck, but if the input is tf.placeholder, it will go through the code successfully but Keras model doesn't allow a non-Input input layer. How can I fix this? Please help me.

The layer is implemented like:

`
class ElmoEmbeddingLayer(Layer):

def __init__(self, config, **kwargs):
self.dimensions = 200
self.options_file = config.options_file
self.weights_file = config.weights_file
self.token_embedding_file = config.token_embedding_file
self.bilm = BidirectionalLanguageModel(
self.options_file,
self.weights_file,
use_character_inputs=False,
embedding_weight_file=self.token_embedding_file,
max_batch_size=1024
)
super(ElmoEmbeddingLayer, self).__init__(**kwargs)

def build(self, input_shape):
super(ElmoEmbeddingLayer, self).build(input_shape)

def call(self, x, mask=None):
context_embeddings_op = self.bilm(x)
elmo_embedding = weight_layers('elmo_output', context_embeddings_op, l2_coef=0.0)
elmo_embedding = elmo_embedding['weighted_op']
return elmo_embedding

def compute_mask(self, inputs, mask=None):
return K.not_equal(inputs, 0)

def compute_output_shape(self, input_shape):
return input_shape[0], input_shape[1], self.dimensions`

And when I use it I just:

`
elmo_model = ElmoEmbeddingLayer(self.data_config)

tmp = tf.placeholder(tf.int32, shape=(None, None))

char_ids = Input(batch_shape=(None, None), dtype='int32', name='input_ids')

elmo_embeddings = elmo_model(char_ids)

lstm_output_1 = Bidirectional(LSTM(units=_char_lstm_size, return_sequences=True))(elmo_embeddings)
`
By using "char_ids", it will be stuck; if I use "tmp", the model can't be formed for regarding non-Input layers as the input of model.