tensorflow saved model to java

Hello data scientists (I'm not), I adapted a multi-class-text-classification-cnn from https://github.com/jiegzhan/multi-class-text-classification-cnn (thanks Jie Zhang) It works ! 90% accuracy ! Now my issue ;-) I really need to load my model with java in order to allow my existing java service to predict categories. According to documentations, I need to use SavedModelBuilder which is the new way of modele saving. I tried for hours but I misse a lot of things... About signature, assets. If someone could help me with that ? => replace the old saver.save by tf.saved_model.builder.SavedModelBuilder into https://github.com/jiegzhan/multi-class-text-classification-cnn/blob/master/train.py He or she will have my infinite gratitude

train.py code:

builder = tf.saved_model.builder.SavedModelBuilder("/home/datam/cnn-text-classification-tf/model/20180423")
sess,
[tf.saved_model.tag_constants.SERVING],
clear_devices=True
)
builder.save()


Java Coding:

SavedModelBundle model = SavedModelBundle.load(SAVED_MODEL_PATH+"/20180423", "serve");
Session s = model.session();
Graph g = model.graph();