Comment by matrix2596

have u seen the dspark paper, they add a morkov head for light weight dependency, not sure whether it can be extended to multi step refining