Vespa Keras weirdness

keras.py

import keras                                                                           
import numpy as np                                                                     
                                                                                       
input_l = keras.Input(shape=(1,), name='input')                                        
layer_1 = keras.layers.Dense(1, activation='relu', name='layer_1')(input_l)            
output_l = keras.layers.Dense(1, activation='linear', name='output')(layer_1)          
model = keras.Model(inputs=[input_l], outputs=[output_l])                              
model.compile(loss='mean_absolute_error', optimizer='rmsprop')                         
                                                                                       
x = np.array(np.arange(1, 100000))                                                     
y = np.array(np.arange(1, 100000))                                                     
                                                                                       
model.fit(np.array(x).reshape(-1),np.array(y), epochs=2, shuffle=False, batch_size=100)
                                                                                       
from keras import backend as K                                                         
import tensorflow as tf                                                                
                                                                                       
print("TF Version: {0}".format(tf.VERSION))                                            
keras.backend.get_session().run(tf.global_variables_initializer())                     
                                                                                       
signature = tf.saved_model.signature_def_utils.predict_signature_def(                  
    inputs={'userid': model.input}, outputs={'scores': model.output})                  
                                                                                       
builder = tf.saved_model.builder.SavedModelBuilder('func_model')                       
builder.add_meta_graph_and_variables(                                                  
    sess=K.get_session(),                                                              
    tags=[tf.saved_model.tag_constants.SERVING],                                       
    signature_def_map={                                                                
        tf.saved_model.signature_constants.DEFAULT_SERVING_SIGNATURE_DEF_KEY:          
            signature                                                                  
    })                                                                                 
builder.save()                                                                         

keras_signature.txt

$ saved_model_cli show --dir func_model/ --all

MetaGraphDef with tag-set: 'serve' contains the following SignatureDefs:

signature_def['serving_default']:
  The given SavedModel SignatureDef contains the following input(s):
    inputs['userid'] tensor_info:
        dtype: DT_FLOAT
        shape: (-1, 1)
        name: input:0
  The given SavedModel SignatureDef contains the following output(s):
    outputs['scores'] tensor_info:
        dtype: DT_FLOAT
        shape: (-1, 1)
        name: output/BiasAdd:0
  Method name is: tensorflow/serving/predict

tfkeras.py

import tensorflow as tf                                                                 
import numpy as np                                                                      
                                                                                        
input_l = tf.keras.Input(shape=(1,), name='input')                                      
layer_1 = tf.keras.layers.Dense(1, activation='relu', name='layer_1')(input_l)          
output_l = tf.keras.layers.Dense(1, activation='linear', name='output')(layer_1)        
model = tf.keras.Model(inputs=[input_l], outputs=[output_l])                            
model.compile(loss='mean_absolute_error', optimizer='rmsprop')                          
                                                                                        
x = np.array(np.arange(1, 100000))                                                      
y = np.array(np.arange(1, 100000))                                                      
                                                                                        
model.fit(np.array(x).reshape(-1),np.array(y), epochs=2, shuffle=False, batch_size=100) 
                                                                                        
print("TF Version: {0}".format(tf.VERSION))                                             
tf.keras.backend.get_session().run(tf.global_variables_initializer())                   
                                                                                        
signature = tf.saved_model.signature_def_utils.predict_signature_def(                   
    inputs={'userid': model.input}, outputs={'scores': model.output})                   
                                                                                        
builder = tf.saved_model.builder.SavedModelBuilder('tfkeras_model')                     
builder.add_meta_graph_and_variables(                                                   
    sess=tf.keras.backend.get_session(),                                                
    tags=[tf.saved_model.tag_constants.SERVING],                                        
    signature_def_map={                                                                 
        tf.saved_model.signature_constants.DEFAULT_SERVING_SIGNATURE_DEF_KEY:           
            signature                                                                   
    })                                                                                  
builder.save()                                                                          

tfkeras_signature.txt

$ saved_model_cli show --dir tfkeras_model/ --all

MetaGraphDef with tag-set: 'serve' contains the following SignatureDefs:

signature_def['serving_default']:
  The given SavedModel SignatureDef contains the following input(s):
    inputs['userid'] tensor_info:
        dtype: DT_FLOAT
        shape: (-1, 1)
        name: input:0
  The given SavedModel SignatureDef contains the following output(s):
    outputs['scores'] tensor_info:
        dtype: DT_FLOAT
        shape: (-1, 1)
        name: output/BiasAdd:0
  Method name is: tensorflow/serving/predict