straight-through-gradient
Relaxit: A Python Library for Optimizing Discrete Probability Distributions in Neural Networks