# EasyRAG: Efficient Retrieval-Augmented Generation Framework for Network Automated Operations

**Authors:** Zhangchi Feng, Dongdong Kuang, Zhongyuan Wang, Zhijie Nie, Yaowei Zheng, Richong Zhang  
**arXiv:** [2410.10315](https://arxiv.org/abs/2410.10315)  
**Date:** 2024-10-14  
**Summary page:** https://www.summarizepaper.com/en/paper/2410.10315v1/

## Key points

- EasyRAG is a simple, lightweight, and efficient framework for network automated operations
- Key advantages include accurate question answering capabilities achieved through specific data processing workflow, dual-route sparse retrieval for coarse ranking, LLM Reranker for reranking, and LLM answer generation and optimization
- EasyRAG secured first place in the GLM4 track during the preliminary round and second place in the semifinals
- Simple deployment features such as BM25 retrieval and BGE-reranker reranking without fine-tuning models
- Efficient inference acceleration scheme reduces inference latency while maintaining high accuracy levels
- Code and data related to EasyRAG are openly available on GitHub at https://github.com/BUAADreamer/EasyRAG

## Summary

, , , , 


In their paper titled "EasyRAG: Efficient Retrieval-Augmented Generation Framework for Network Automated Operations," authors Zhangchi Feng, Dongdong Kuang, Zhongyuan Wang, Zhijie Nie, Yaowei Zheng, and Richong Zhang introduce EasyRAG as a simple, lightweight, and efficient framework for network automated operations. <break>

The key advantages of EasyRAG include its accurate question answering capabilities achieved through a specific data processing workflow, dual-route sparse retrieval for coarse ranking, LLM Reranker for reranking, and LLM answer generation and optimization. Notably, this approach secured first place in the GLM4 track during the preliminary round and second place in the semifinals. <break>

Furthermore, EasyRAG boasts simple deployment features such as BM25 retrieval and BGE-reranker reranking without the need for fine-tuning models. It occupies minimal VRAM, is easy to deploy, highly scalable, and offers a flexible code library with various search and generation strategies to facilitate custom process implementation. <break>

The authors also designed an efficient inference acceleration scheme that significantly reduces the inference latency of RAG while maintaining high accuracy levels. This scheme can be seamlessly integrated into any component of the RAG process to enhance overall system efficiency. <break>

The code and data related to EasyRAG are openly available on GitHub at https://github.com/BUAADreamer/EasyRAG. Overall,<break> 
EasyRAG presents a comprehensive solution for network automated operations by combining accurate question answering capabilities with simple deployment features and efficient inference acceleration schemes.