PACE-RAG

Evidence frontloading and pressure-adaptive budgeting for RAG.

Overview

PACE-RAG studies how the bottleneck in retrieval-augmented generation can shift between reranking and LLM generation. It combines evidence frontloading with a reranking budget that adapts to serving pressure, aiming to retain useful evidence while reducing latency.

PACE-RAG overview.

BibTeX

1. @article{cai2026less,
     title = {Less can be More: Relieving RAG Bottlenecks via Evidence Frontloading and Pressure-Adaptive Budgeting},
     author = {Cai, Weibin and Zafarani, Reza},
     journal = {arXiv preprint arXiv:2608.25115},
     year = {2026},
     keywords = {conference},
   }