Question Answering on HotpotQA (ATR, ASR, Accuracy)
91.3AccuracyCleanRAG
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| CleanRAGLLM=GPT-3.52025.08 | 91.3 | 0 | 8 | |
| CleanRAGLLM=GPT-42025.08 | 90.8 | 0 | 9 | |
| CleanRAGLLM=GPT-4o2025.08 | 90.7 | 0 | 9 | |
| ML-FilterRAGLLM=GPT-42025.08 | 90.5 | 1.5 | 9.1 | |
| ML-FilterRAGLLM=GPT-3.52025.08 | 90.3 | 1.5 | 9 | |
| ML-FilterRAGLLM=GPT-4o2025.08 | 89.9 | 1.5 | 9.4 | |
| FilterRAGepsilon=0.2, LLM=GPT-3.52025.08 | 88.1 | 0 | 8.2 | |
| FilterRAGepsilon=0.2, LLM=GPT-42025.08 | 87 | 0 | 9 | |
| FilterRAGepsilon=0.2, LLM=GPT-4o2025.08 | 82 | 0 | 9.1 | |
| CleanRAGLLM=LLaMA-32025.08 | 66.5 | 0 | 31 | |
| Neuro-RITRanking Strategy=Reranker2026.04 | 55.71 | — | — | |
| PA-RAGRanking Strategy=Reranker2026.04 | 54.12 | — | — | |
| ML-FilterRAGLLM=LLaMA-32025.08 | 54.1 | 1.5 | 33 | |
| PA-RAGRanking Strategy=RankCoT2026.04 | 53.89 | — | — | |
| FilterRAGepsilon=0.2, LLM=LLaMA-32025.08 | 53.6 | 0 | 38 | |
| Neuro-RIT + RerankerGenerator=Mistral-7B-Instruct-v0.22026.04 | 53.25 | — | — | |
| Neuro-RIT + RankCoTGenerator=Mistral-7B-Instruct-v0.22026.04 | 52.45 | — | — | |
| Neuro-RITRanking Strategy=None2026.04 | 52.25 | — | — | |
| Neuro-RITRanking Strategy=RankCoT2026.04 | 52.04 | — | — | |
| InstructRAGRanking Strategy=RankCoT2026.04 | 51.66 | — | — | |
| RALM + RerankerGenerator=Mistral-7B-Instruct-v0.22026.04 | 50.62 | — | — | |
| Neuro-RITGenerator=Mistral-7B-Instruct-v0.22026.04 | 50.46 | — | — | |
| RALM + RankCoTGenerator=Mistral-7B-Instruct-v0.22026.04 | 49.73 | — | — | |
| PA-RAGRanking Strategy=None2026.04 | 49.29 | — | — | |
| RALMRanking Strategy=RankCoT2026.04 | 48.54 | — | — | |
| RALMGenerator=Mistral-7B-Instruct-v0.22026.04 | 48.39 | — | — | |
| InstructRAGRanking Strategy=Reranker2026.04 | 48.32 | — | — | |
| RALMRanking Strategy=None2026.04 | 45.44 | — | — | |
| InstructRAGRanking Strategy=None2026.04 | 45.33 | — | — | |
| RALMRanking Strategy=Reranker2026.04 | 45.27 | — | — | |
| RetRobustRanking Strategy=RankCoT2026.04 | 43.84 | — | — | |
| RetRobustRanking Strategy=Reranker2026.04 | 41.46 | — | — | |
| RetRobustRanking Strategy=None2026.04 | 40.84 | — | — | |
| Baseline (CAD)Model=Llama3-8B2026.04 | 39.5 | — | — | |
| Token-aware CFBModel=Llama3-8B2026.04 | 37.1 | — | — | |
| Context-aware CFBModel=Llama3-8B2026.04 | 36.8 | — | — | |
| Static CFBModel=Llama3-8B2026.04 | 34.7 | — | — | |
| PoisonedRAGLLM=GPT-3.52025.08 | 4.2 | 100 | 94 | |
| PoisonedRAGLLM=LLaMA-32025.08 | 0.1 | 100 | 98 | |
| PoisonedRAGLLM=GPT-42025.08 | 0 | 100 | 90 | |
| PoisonedRAGLLM=GPT-4o2025.08 | 0 | 100 | 93 |