Question Answering on RQA (poison @ Position 1, k=10) (test)
0.7Robustness AccuracyMIS
Evaluation Results
| Method | Links | |
|---|---|---|
| MISModel=Mistral-7B, k=102025.09 | 0.7 | |
| MISModel=GPT-4o-mini, k=102025.09 | 0.7 | |
| MISModel=Llama3.2-3B, k=102025.09 | 0.67 | |
| RobustRAGModel=GPT-4o-mini, k=102025.09 | 0.67 | |
| RobustRAGModel=Llama3.2-3B, k=102025.09 | 0.61 | |
| RobustRAGModel=Mistral-7B, k=102025.09 | 0.53 | |
| Vanilla RAGModel=Llama3.2-3B, k=102025.09 | 0.51 | |
| Vanilla RAGModel=GPT-4o-mini, k=102025.09 | 0.45 | |
| Vanilla RAGModel=Mistral-7B, k=102025.09 | 0.43 | |
| InstructRAGModel=Mistral-7B, k=102025.09 | 0.4 | |
| AstuteRAGModel=GPT-4o-mini, k=102025.09 | 0.39 | |
| InstructRAGModel=GPT-4o-mini, k=102025.09 | 0.37 | |
| AstuteRAGModel=Llama3.2-3B, k=102025.09 | 0.36 | |
| InstructRAGModel=Llama3.2-3B, k=102025.09 | 0.3 | |
| AstuteRAGModel=Mistral-7B, k=102025.09 | 0.24 |