HiFi-RAG: Improved RAG for Open-Domain QA
Published:Dec 27, 2025 02:37
•1 min read
•ArXiv
Analysis
This paper presents HiFi-RAG, a novel Retrieval-Augmented Generation (RAG) system that won the MMU-RAGent NeurIPS 2025 competition. The core innovation lies in a hierarchical filtering approach and a two-pass generation strategy leveraging different Gemini 2.5 models for efficiency and performance. The paper highlights significant improvements over baselines, particularly on a custom dataset focusing on post-cutoff knowledge, demonstrating the system's ability to handle recent information.
Key Takeaways
- •HiFi-RAG is a novel RAG system employing hierarchical filtering and two-pass generation.
- •It leverages Gemini 2.5 Flash for efficiency and Gemini 2.5 Pro for reasoning.
- •The system achieves significant performance gains, especially on post-cutoff knowledge tasks.
- •The approach demonstrates the effectiveness of multi-stage pipelines in RAG.
Reference
“HiFi-RAG outperforms the parametric baseline by 57.4% in ROUGE-L and 14.9% in DeBERTaScore on Test2025.”