CoinRAG：情境化資訊片段的KV快取重用於長上下文RAG

摘要

近期針對檢索增強生成（RAG）的最佳化研究利用區塊層級的 KV 快取重用，以避免處理冗長的檢索上下文以提升效率，然而粗粒度區塊中仍存在顯著的資訊冗餘與雜訊。本文提出 CoinRAG（Contextualized Information Nugget KV Cache Reuse for Long-Context RAG），在低預填充延遲限制下最佳化柏拉圖效率前沿，同時最大化準確度。此名稱以比喻方式反映我們的核心機制：如同將小額代幣（或「硬幣」）組裝以累積更大的價值，CoinRAG 以組合方式重用離線計算的細粒度資訊碎片快取，以更具語意相關性且更精簡的方式高效率地形成學習式上下文表徵。具體而言，CoinRAG 不進行完整區塊編碼，而是透過兩階段檢索識別檢索區塊中與查詢相關的語意單元，並將其切片後的 KV 表徵與區塊層級上下文無縫組裝。在 LongBench 多跳問答任務上的廣泛評估表明，CoinRAG 顯著降低營運成本，並在標準快速預填充延遲預算下，以新的柏拉圖效率前沿和平均 5.3% 的答案品質（F1）相對提升優於其他基準方法。

English

Recent optimization studies on Retrieval-Augmented Generation (RAG) have exploited chunk-level KV cache reuse to avoid processing long retrieved contexts for higher efficiency, while significant information redundancy and noise still remain in the coarse-grained chunks. This paper optimizes the Pareto frontier under low prefill latency constraints while maximizing accuracy by proposing CoinRAG (Contextualized Information Nugget KV Cache Reuse for Long-Context RAG). The name metaphorically reflects our core mechanism: much like assembling small tokens (or "coins") to accumulate a larger value, CoinRAG compositionally reuses offline-computed, fine-grained nugget caches to form a learned contextual representation efficiently in a more semantically relevant but compact manner. Specifically, instead of full-chunk encoding, CoinRAG identifies query-relevant semantic units within retrieved chunks through two-stage retrieval and seamlessly assembles their sliced KV representations with a chunk-level context. Extensive evaluations on LongBench multi-hop question answering tasks demonstrate that CoinRAG significantly reduces operational costs and outperforms the other baselines with a new Pareto frontier and an average 5.3% relative improvement in answer quality (F1) under a standard fast prefill latency budget.