Question Answering on NQ poison @ Position 10, k=10 (test)
65.8Robustness AccuracyMIS
Evaluation Results
| Method | Links | |
|---|---|---|
| MISModel=GPT-4o-mini, k=102025.09 | 65.8 | |
| Vanilla RAGModel=GPT-4o-mini, k=102025.09 | 63.6 | |
| MISModel=Llama3.2-3B, k=102025.09 | 60.4 | |
| RobustRAGModel=GPT-4o-mini, k=102025.09 | 58.8 | |
| MISModel=Mistral-7B, k=102025.09 | 58 | |
| AstuteRAGModel=GPT-4o-mini, k=102025.09 | 56.8 | |
| InstructRAGModel=GPT-4o-mini, k=102025.09 | 54.8 | |
| RobustRAGModel=Llama3.2-3B, k=102025.09 | 52.8 | |
| InstructRAGModel=Mistral-7B, k=102025.09 | 52 | |
| AstuteRAGModel=Llama3.2-3B, k=102025.09 | 51 | |
| AstuteRAGModel=Mistral-7B, k=102025.09 | 50.6 | |
| InstructRAGModel=Llama3.2-3B, k=102025.09 | 47 | |
| RobustRAGModel=Mistral-7B, k=102025.09 | 44.4 | |
| Vanilla RAGModel=Llama3.2-3B, k=102025.09 | 42.8 | |
| Vanilla RAGModel=Mistral-7B, k=102025.09 | 41.6 |