Small Language Model framework to score, re-rank, and prune LLM candidates to slash token overhead.