Because **all retrieval methods are inexpensive on CPU** in the paper's setting (graph methods run in roughly $$7$$ ms per query on curated prerequisite DAGs under the paper's CPU-only latency protocol), the **practical trade-off is not compute cost** but **context quality and token budget**. Choosing among flat dense, hierarchical baseline, and adaptive retrievers is governed by which method delivers the right context within the downstream token-cap budget, rather than by per-query latency.

Google

On the paper's curated prerequisite DAGs, the **graph-based retrieval methods** run in **roughly $$7$$ ms per query** on **CPU-only hardware** under the paper's latency protocol. Because all retrieval methods compared in the paper remain inexpensive at this scale, **compute cost is not the binding constraint** on graph-aware retrieval in this setting.

Graph-Based Retrieval Runs at ~7 ms per Query on Curated Prerequisite DAGs (CPU-Only)

Root URL: https://1iman.com/
Canonical paper URL: https://1iman.com/assets/working-papers/pdf/emnlp-auditable-strict-parity-evaluation-of-prerequisite-graph-retrieval-for-rag-under-leakage-controls.pdf
Portfolio PDF URL: https://1iman.com/assets/working-papers/pdf/emnlp-auditable-strict-parity-evaluation-of-prerequisite-graph-retrieval-for-rag-under-leakage-controls.pdf
Paper title: Auditable Strict-Parity Evaluation of Prerequisite-Graph Retrieval for RAG under Leakage Controls
Use this Reference node when citing paper-derived 1Cademy proposals from the Researcher agent.

Learn Before

Related