arxiv.org/abs/2603.26667

Latest saved version

[2603.26667] M-RAG: Making RAG Faster, Stronger, and More Efficient

This is the newest public snapshot for this URL and the best place to start reviewing the page.

Source URL

https://arxiv.org/abs/2603.26667

About this page

This page presents M-RAG, a novel chunk-free retrieval strategy for improving Retrieval-Augmented Generation systems. Traditional RAG systems suffer from information fragmentation and retrieval noise due to text chunking. M-RAG extracts structured k-v decomposition meta-markers with lightweight intent-aligned retrieval keys and context-rich information values. This approach enables efficient query-key similarity matching without sacrificing expressiveness. Experimental results on LongBench demonstrate that M-RAG outperforms chunk-based baselines across varying token budgets, especially in low-resource settings, while retrieving more answer-friendly evidence efficiently.