RankRAG: Unifying Context Ranking with Retrieval-Augmented Generation in LLMs

Benchmark Model Rank Results
question-answering-on-natural-questionsRankRAG-llama3-70b (Zero-Shot, DPR)EM: 50.0
question-answering-on-natural-questionsRankRAG-llama3-70b (Zero-Shot, KILT)EM: 54.2
question-answering-on-natural-questionsRankRAG-llama3-8b (Zero-Shot, DPR)EM: 46.1
question-answering-on-natural-questionsRankRAG-llama3-8b (Zero-Shot, KILT)EM: 50.6
question-answering-on-pubmedqaRankRAG-llama3-70B (Zero-Shot)Accuracy: 79.8
question-answering-on-triviaqaRankRAG-llama3-70b (Zero-Shot, DPR)EM: 72.6
question-answering-on-triviaqaRankRAG-llama3-70b (Zero-Shot, KILT)EM: 86.5
question-answering-on-triviaqaRankRAG-llama3-8b (Zero-Shot, KILT)EM: 82.9