import tensorflow as tf
graph = tf.Graph()
builder = tf.saved_model.builder.SavedModelBuilder('./model')
writer = tf.summary.FileWriter('.')
with graph.as_default():
  a = tf.constant(2, name='a')
  b = tf.constant(3, name='b')
  x = tf.placeholder(tf.int32, name='x')
  y = tf.placeholder(tf.int32, name='y')
  z = tf.math.add(a*x, b*y, name='z')
  writer.add_graph(tf.get_default_graph())
  writer.flush()
  sess = tf.Session()
  sess.run(z, feed_dict = {x: 2, y: 3})
  builder.add_meta_graph_and_variables(sess, [tf.saved_model.tag_constants.SERVING])
  builder.save()