CompRank: Efficient LLM Reranking via Token-Level Compression and Decoding-Free Scoring
Researchers have developed CompRank, a new framework designed to make large language model (LLM) rerankers more computationally efficient for information retrieval tasks. CompRank achieves this by reducing redundant computations through token-level compression and a decoding-free scoring method. Experiments demonstrate that CompRank significantly speeds up reranking while maintaining high performance, making LLM-based reranking more scalable for long candidate lists. AI
IMPACT This research offers a more efficient method for LLM reranking, potentially enabling wider adoption in retrieval systems.