You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This is official version of KGQAGen framework. It interactively generates challenging KGQA instances from seed Wikidata entities using an LLM agent and SPARQL querying. Our KGQAGen-10k example dataset is available in https://huggingface.co/datasets/lianglz/KGQAGen-10k.
QAFD-RAG is a graph-based retrieval-augmented generation framework that uses query-aware flow diffusion to retrieve contextually relevant subgraphs from a knowledge graph with statistical retrieval guarantees.
Reproducible Agentic RAG baselines for multi-hop QA and retrieval: AceSearcher, Search-R1, IRCoT, Self-RAG, and more, with inference and evaluation tools.
Multi-hop QA system combining HopRAG graph traversal with Self-RAG critique gates. Beats HopRAG on HotpotQA, MuSiQue, and 2WikiMultiHopQA. Phase 2 integrates a trained reranker-slm adapter at the IsREL gate for further gains.