Code for our paper:
Leveraging Relaxed Equilibrium by Lazy Transition for Sequence Modeling
#################IMPORTANT#######################################
I've changed the framework of UNIVERSAL this year. For now, lazy_transformer was based on previous framework. I will reimplement this work ASAP.
However, you can still use the lt.py and lazyTransition.py for your Universal Transformer/ Naive Transformer to implement Lazy Transformer.