Comment by nextos

I have applied SATs with some success to generative models where sampling methods would not work well, e.g. if you have many soft and hard constraints. Another alternative is to use neural guides.