RAG Robustness on RQA-MC
80Paradox RACCAV Filter
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| AV FilterLLM=Mistral-7B, Defense=AV Filter, alpha=infinity2025.06 | 80 | 3.4 | 77.2 | 5.2 | 74.6 | 7.6 | |
| AV FilterLLM=Llama2-C, Defense=AV Filter, alpha=infinity2025.06 | 77.6 | 8.2 | 77.2 | 6.8 | 52.6 | 18.8 | |
| AV FilterLLM=Mistral-7B, Defense=AV Filter, alpha=102025.06 | 77.4 | 6.8 | 73.6 | 9 | 68.2 | 11 | |
| AV FilterLLM=Mistral-7B, Defense=AV Filter, alpha=52025.06 | 76 | 8.4 | 74 | 8.6 | 69.4 | 10.8 | |
| AV FilterLLM=Llama2-C, Defense=AV Filter, alpha=102025.06 | 64.2 | 20.8 | 71.2 | 13.8 | 49 | 34.8 | |
| AV FilterLLM=GPT-4o, Defense=AV Filter, alpha=infinity2025.06 | 62.8 | 2.4 | 59.6 | 4.2 | 55.2 | 3 | |
| AV FilterLLM=Llama2-C, Defense=AV Filter, alpha=52025.06 | 59 | 25.4 | 71 | 12.8 | 45 | 39.8 | |
| AV FilterLLM=GPT-4o, Defense=AV Filter, alpha=102025.06 | 58 | 3.8 | 51 | 10.2 | 54.8 | 3.2 | |
| AV FilterLLM=GPT-4o, Defense=AV Filter, alpha=52025.06 | 56.8 | 5.6 | 50.4 | 11 | 54.2 | 3.6 | |
| Vanilla RAGLLM=Mistral-7B, Defense=Vanilla2025.06 | 54.2 | 41.4 | 60.4 | 31.8 | 52 | 32 | |
| Vanilla RAGLLM=Llama2-C, Defense=Vanilla2025.06 | 50 | 42.8 | 57 | 34.2 | 37 | 51.8 | |
| Vanilla RAGLLM=GPT-4o, Defense=Vanilla2025.06 | 31.6 | 37.2 | 41 | 25 | 53.4 | 4.6 |