model_architecture.py

    x = tf.keras.layers.Dense(256, activation='relu')(pooled_output)
    dense = tf.keras.layers.Dense(64, activation='relu')(x)
    pred = tf.keras.layers.Dense(1, activation='sigmoid')(dense)
 
    model = tf.keras.Model(
        inputs=[inputs['input_word_ids'], 
                inputs['input_mask'], 
                inputs['input_type_ids']], 
        outputs=pred
    )
    model.compile(loss='binary_crossentropy', 
                        optimizer='adam', 
                        metrics=['accuracy'])