DEEP-PolyU/LinearRAG
Source code of LinearRAG at ICLR'26
Constructs knowledge graphs without LLM-based relation extraction, instead using lightweight entity recognition and semantic embeddings for linking—eliminating token costs while maintaining linear time/space complexity. Enables multi-hop reasoning through semantic bridging in a single retrieval pass, achieving competitive performance on complex QA benchmarks without explicit relational graphs. Integrates Spacy for NLP processing, sentence transformers for embeddings, and supports GPU-accelerated vectorized retrieval alongside BFS-based lookups.
421 stars.
Stars
421
Forks
45
Language
Python
License
GPL-3.0
Category
Last pushed
Mar 04, 2026
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/rag/DEEP-PolyU/LinearRAG"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Higher-rated alternatives
LearningCircuit/local-deep-research
Local Deep Research achieves ~95% on SimpleQA benchmark (tested with GPT-4.1-mini). Supports...
NVIDIA-AI-Blueprints/rag
This NVIDIA RAG blueprint serves as a reference solution for a foundational Retrieval Augmented...
Denis2054/RAG-Driven-Generative-AI
This repository provides programs to build Retrieval Augmented Generation (RAG) code for...
0verL1nk/PaperSage
đź“š AI-powered research reading workbench. Project-based paper Q&A with Hybrid RAG, multi-agent...
RapidFireAI/rapidfireai
RapidFire AI: Rapid AI Customization from RAG to Fine-Tuning