Freezing model for production

I am trying to freeze the trained net model produced with train.py for production in flask. However I am struggling to do this, since then freezing operation is asking for "output-nodes" and I am not entirely sure which ones I should be using.  Can anyone help?