RAG Robustness on RQA
66.4Paradox RACCAV Filter
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| AV FilterLLM=GPT-4o, Defense=AV Filter, alpha=infinity2025.06 | 66.4 | 4.6 | 68.2 | 2.4 | 65.2 | 5.2 | |
| AV FilterLLM=Mistral-7B, Defense=AV Filter, alpha=infinity2025.06 | 65.2 | 3.8 | 65.6 | 2.8 | 61.8 | 7.4 | |
| AV FilterLLM=GPT-4o, Defense=AV Filter, alpha=102025.06 | 65.2 | 6 | 58 | 8.6 | 65.6 | 6 | |
| AV FilterLLM=GPT-4o, Defense=AV Filter, alpha=52025.06 | 63.2 | 7.6 | 57.4 | 9.4 | 66 | 6.8 | |
| AV FilterLLM=Llama2-C, Defense=AV Filter, alpha=infinity2025.06 | 59.8 | 1.8 | 59.2 | 2.4 | 45.2 | 38.8 | |
| AV FilterLLM=Mistral-7B, Defense=AV Filter, alpha=102025.06 | 59.4 | 4.8 | 54 | 7.8 | 59.2 | 8 | |
| AV FilterLLM=Llama2-C, Defense=AV Filter, alpha=102025.06 | 58.4 | 6.2 | 55 | 7.2 | 53 | 17.4 | |
| AV FilterLLM=Llama2-C, Defense=AV Filter, alpha=52025.06 | 58 | 6.6 | 54.4 | 5.8 | 54 | 15.4 | |
| AV FilterLLM=Mistral-7B, Defense=AV Filter, alpha=52025.06 | 57.6 | 4.2 | 54 | 8 | 57.2 | 7.2 | |
| Vanilla RAGLLM=GPT-4o, Defense=Vanilla2025.06 | 41.2 | 45 | 48.4 | 33 | 61.2 | 11.6 | |
| Vanilla RAGLLM=Llama2-C, Defense=Vanilla2025.06 | 37.2 | 39.8 | 50.2 | 22 | 20 | 63.4 | |
| Vanilla RAGLLM=Mistral-7B, Defense=Vanilla2025.06 | 30.4 | 35 | 39.4 | 26.6 | 43.6 | 17.8 |