tensorflow，keras，模型的输出张量必须是TensorFlow`Layer`的输出

Question

第一个模型因ValueError: Output tensors to a Model must be the output of a TensorFlow层(thus holding past layer metadata). Found: Tensor("dense/BiasAdd:0", shape=(?, 1, 1), dtype=float32)而出错

import tensorflow as tf
from tensorflow.keras.layers import Input, Embedding, concatenate, Dense, Flatten

 def get_model():
     first_input = Input(shape = (1,), name='first_input')
     second_input = Input(shape = (1,), name='second_input' )

     embedding_layer = Embedding(input_dim=10, output_dim=3, input_length=1)
     first_input_encoded = embedding_layer(first_input)
     second_input_encoded = embedding_layer(second_input)

     o = concatenate([first_input_encoded, second_input_encoded])
     o = Dense(1)(o)                                                                                                                                                                  
     model = tf.keras.models.Model(inputs=[first_input, second_input_encoded], outputs=o)
     return model

第二个方法有效，但我不理解该错误的含义。

def get_model():                                                                                                                                     
    first_input = Input(shape = (1,), name='first_input')                                                                                                         
    second_input = Input(shape = (5,), name='second_input' )

    embedding_layer = Embedding(input_dim=10, output_dim=3, input_length=1)
    first_input_encoded = embedding_layer(first_input)
    second_input_encoded = embedding_layer(second_input)

    second = tf.reduce_mean(second_input_encoded, axis=1, keepdims=True)                                                                                                                                                                                                                                                        
    o = concatenate([first_input_encoded, second])                                                                                                                
    o = Dense(1)(o)                                                                                                                                                                                                                                                                                                             
    model = tf.keras.models.Model(inputs=[first_input, second_input], outputs=o)                                                                                  
    return model