Question Answering on NQ (ACCR and ACCL)
87AccuracyHRAG
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| HRAGModel=Llama3-70B, Defense Framework=AstuteRAG2025.05 | 87 | — | — | 5 | — | |
| CleanModel=Llama3-70B, Defense Framework=AstuteRAG2025.05 | 86 | — | — | — | — | |
| PRAGModel=Deepseek-V3-0324, Defense Framework=AstuteRAG2025.05 | 86 | — | — | 0 | — | |
| CleanModel=Deepseek-V3-0324, Defense Framework=AstuteRAG2025.05 | 86 | — | — | — | — | |
| CleanModel=GPT-4o, Defense Framework=AstuteRAG2025.05 | 84 | — | — | — | — | |
| HRAGModel=Deepseek-V3-0324, Defense Framework=AstuteRAG2025.05 | 84 | — | — | 6 | — | |
| PRAGModel=GPT-4o, Defense Framework=InstructRAG2025.05 | 83 | — | — | 14 | — | |
| HRAGModel=GPT-4, Defense Framework=AstuteRAG2025.05 | 83 | — | — | 7 | — | |
| PRAGModel=Llama3-70B, Defense Framework=AstuteRAG2025.05 | 83 | — | — | 5 | — | |
| CleanModel=Deepseek-V3-0324, Defense Framework=InstructRAG2025.05 | 82 | — | — | — | — | |
| HRAGModel=GPT-4o, Defense Framework=AstuteRAG2025.05 | 82 | — | — | 9 | — | |
| PRAGModel=GPT-4, Defense Framework=AstuteRAG2025.05 | 81 | — | — | 5 | — | |
| CleanModel=GPT-4, Defense Framework=AstuteRAG2025.05 | 80 | — | — | — | — | |
| PRAGModel=GPT-4o, Defense Framework=AstuteRAG2025.05 | 80 | — | — | 8 | — | |
| CleanModel=Llama3-70B, Defense Framework=InstructRAG2025.05 | 79 | — | — | — | — | |
| PRAGModel=Deepseek-V3-0324, Defense Framework=InstructRAG2025.05 | 79 | — | — | 21 | — | |
| CleanModel=GPT-4o, Defense Framework=InstructRAG2025.05 | 78 | — | — | — | — | |
| PRAGModel=Llama3-8B, Defense Framework=AstuteRAG2025.05 | 78 | — | — | 9 | — | |
| CleanModel=Llama3-8B, Defense Framework=AstuteRAG2025.05 | 78 | — | — | — | — | |
| CleanModel=GPT-4, Defense Framework=InstructRAG2025.05 | 75 | — | — | — | — | |
| HRAGModel=GPT-4o, Defense Framework=InstructRAG2025.05 | 75 | — | — | 18 | — | |
| HRAGModel=GPT-3.5, Defense Framework=AstuteRAG2025.05 | 75 | — | — | 9 | — | |
| AuthChainModel=Deepseek-V3-0324, Defense Framework=AstuteRAG2025.05 | 75 | — | — | 11 | — | |
| CleanModel=Llama3-8B, Defense Framework=InstructRAG2025.05 | 74 | — | — | — | — | |
| DCRDModel=Llama3-8B2026.05 | 73.4 | — | — | — | — | |
| HRAGModel=Llama3-70B, Defense Framework=InstructRAG2025.05 | 72 | — | — | 16 | — | |
| DCRDModel=Llama2-13B2026.05 | 71.4 | — | — | — | — | |
| CleanModel=GPT-3.5, Defense Framework=AstuteRAG2025.05 | 71 | — | — | — | — | |
| AuthChainModel=GPT-4o, Defense Framework=AstuteRAG2025.05 | 71 | — | — | 23 | — | |
| AuthChainModel=GPT-4o, Defense Framework=InstructRAG2025.05 | 69 | — | — | 27 | — | |
| HRAGModel=Deepseek-V3-0324, Defense Framework=InstructRAG2025.05 | 69 | — | — | 28 | — | |
| AuthChainModel=GPT-4, Defense Framework=AstuteRAG2025.05 | 69 | — | — | 23 | — | |
| HRAGModel=Llama3-8B, Defense Framework=AstuteRAG2025.05 | 69 | — | — | 17 | — | |
| COIECDModel=Llama2-13B2026.05 | 68.6 | — | — | — | — | |
| DCRDModel=Llama2-7B2026.05 | 68.4 | — | — | — | — | |
| COIECDModel=Llama3-8B2026.05 | 68.4 | — | — | — | — | |
| ADACADModel=Llama2-13B2026.05 | 68.2 | — | — | — | — | |
| AuthChainModel=Deepseek-V3-0324, Defense Framework=InstructRAG2025.05 | 68 | — | — | 30 | — | |
| GreedyModel=Llama3-8B2026.05 | 67.4 | — | — | — | — | |
| CleanModel=GPT-3.5, Defense Framework=InstructRAG2025.05 | 66 | — | — | — | — | |
| ADACADModel=Llama2-7B2026.05 | 65.4 | — | — | — | — | |
| ADACADModel=Llama3-8B2026.05 | 65.2 | — | — | — | — | |
| PRAGModel=Llama3-70B, Defense Framework=InstructRAG2025.05 | 65 | — | — | 30 | — | |
| GreedyModel=Llama2-13B2026.05 | 64.3 | — | — | — | — | |
| PRAGModel=GPT-4, Defense Framework=InstructRAG2025.05 | 64 | — | — | 29 | — | |
| RADARTop-k=10, Pos=Pos 10, Generator LLM=Grok-4-fast2026.05 | 62.4 | — | — | 5 | — | |
| PRAGModel=GPT-3.5, Defense Framework=AstuteRAG2025.05 | 62 | — | — | 21 | — | |
| ReliabilityRAGTop-k=50, Pos=Pos 25, Generator LLM=Grok-4-fast2026.05 | 62 | — | — | 3.6 | — | |
| RADARTop-k=50, Pos=Pos 25, Generator LLM=Grok-4-fast2026.05 | 62 | — | — | 4 | — | |
| RobustRAGTop-k=50, Pos=Pos 50, Generator LLM=Grok-4-fast2026.05 | 62 | — | — | 3.2 | — | |
| ReliabilityRAGTop-k=50, Pos=Pos 50, Generator LLM=Grok-4-fast2026.05 | 62 | — | — | 4 | — | |
| RADARTop-k=50, Pos=Pos 50, Generator LLM=Grok-4-fast2026.05 | 62 | — | — | 4 | — | |
| ReliabilityRAGTop-k=10, Pos=Pos 10, Generator LLM=Grok-4-fast2026.05 | 61.4 | — | — | 9.8 | — | |
| RobustRAGTop-k=50, Pos=Pos 25, Generator LLM=Grok-4-fast2026.05 | 61.2 | — | — | 2.8 | — | |
| RADARTop-k=50, Pos=Pos 1, Generator LLM=Grok-4-fast2026.05 | 61 | — | — | 4.8 | — | |
| DCRDModel=Mistral-7B2026.05 | 60.9 | — | — | — | — | |
| AuthChainModel=Llama3-8B, Defense Framework=AstuteRAG2025.05 | 60 | — | — | 33 | — | |
| COIECDModel=Llama2-7B2026.05 | 59.9 | — | — | — | — | |
| RADARTop-k=10, Pos=Pos 1, Generator LLM=Grok-4-fast2026.05 | 59.4 | — | — | 8.2 | — | |
| COIECDModel=Mistral-7B2026.05 | 59.1 | — | — | — | — | |
| GreedyModel=Mistral-7B2026.05 | 58.8 | — | — | — | — | |
| ADACADModel=Mistral-7B2026.05 | 58.1 | — | — | — | — | |
| AuthChainModel=Llama3-70B, Defense Framework=AstuteRAG2025.05 | 58 | — | — | 38 | — | |
| RobustRAGTop-k=50, Pos=Pos 1, Generator LLM=Grok-4-fast2026.05 | 58 | — | — | 5.2 | — | |
| ReliabilityRAGTop-k=10, Pos=Pos 1, Generator LLM=Grok-4-fast2026.05 | 57 | — | — | 16.8 | — | |
| RobustRAGTop-k=10, Pos=Pos 10, Generator LLM=Grok-4-fast2026.05 | 57 | — | — | 2.2 | — | |
| ILRBackbone=Llama-3.1-8B-Instruct, Training Group=Group 22025.09 | 55.31 | — | — | — | — | |
| ReliabilityRAGTop-k=50, Pos=Pos 1, Generator LLM=Grok-4-fast2026.05 | 55 | — | — | 25.4 | — | |
| ILRBackbone=Llama-3.1-8B-Instruct, Training Group=Group 12025.09 | 54.77 | — | — | — | — | |
| RobustRAGTop-k=10, Pos=Pos 1, Generator LLM=Grok-4-fast2026.05 | 54.4 | — | — | 5.2 | — | |
| HRAGModel=GPT-4, Defense Framework=InstructRAG2025.05 | 54 | — | — | 40 | — | |
| PRAGModel=Llama3-8B, Defense Framework=InstructRAG2025.05 | 54 | — | — | 38 | — | |
| ILRBackbone=Qwen2.5-14B-Instruct, Training Group=Group 32025.09 | 53.87 | — | — | — | — | |
| CADModel=Mistral-7B2026.05 | 52.6 | — | — | — | — | |
| ILRBackbone=Qwen2.5-14B-Instruct, Training Group=Group 22025.09 | 52.45 | — | — | — | — | |
| AstuteRAGTop-k=10, Pos=Pos 10, Generator LLM=Grok-4-fast2026.05 | 52.4 | — | — | 2.6 | — | |
| DPOBackbone=Llama-3.1-8B-Instruct2025.09 | 52.32 | — | — | — | — | |
| AstuteRAGTop-k=10, Pos=Pos 1, Generator LLM=Grok-4-fast2026.05 | 52.2 | — | — | 2.6 | — | |
| GRPOBackbone=Llama-3.1-8B-Instruct2025.09 | 52.11 | — | — | — | — | |
| GreedyModel=Llama2-7B2026.05 | 51.9 | — | — | — | — | |
| InstructRAGTop-k=50, Pos=Pos 25, Generator LLM=Grok-4-fast2026.05 | 51.6 | — | — | 3.6 | — | |
| CADModel=Llama2-13B2026.05 | 51.3 | — | — | — | — | |
| BaseBackbone=Llama-3.1-8B-Instruct2025.09 | 51.12 | — | — | — | — | |
| AuthChainModel=GPT-4, Defense Framework=InstructRAG2025.05 | 51 | — | — | 44 | — | |
| CADModel=Llama3-8B2026.05 | 51 | — | — | — | — | |
| InstructRAGTop-k=50, Pos=Pos 50, Generator LLM=Grok-4-fast2026.05 | 50.4 | — | — | 4 | — | |
| CADModel=Llama2-7B2026.05 | 50.3 | — | — | — | — | |
| HRAGModel=Llama3-8B, Defense Framework=InstructRAG2025.05 | 50 | — | — | 40 | — | |
| GRPOBackbone=Qwen2.5-14B-Instruct2025.09 | 50 | — | — | — | — | |
| AstuteRAGTop-k=50, Pos=Pos 25, Generator LLM=Grok-4-fast2026.05 | 50 | — | — | 3.8 | — | |
| AstuteRAGTop-k=50, Pos=Pos 50, Generator LLM=Grok-4-fast2026.05 | 49.8 | — | — | 3 | — | |
| DPOBackbone=Qwen2.5-14B-Instruct2025.09 | 49.6 | — | — | — | — | |
| AstuteRAGTop-k=50, Pos=Pos 1, Generator LLM=Grok-4-fast2026.05 | 49.2 | — | — | 2.2 | — | |
| BaseBackbone=Qwen2.5-14B-Instruct2025.09 | 49.13 | — | — | — | — | |
| PRAGModel=GPT-3.5, Defense Framework=InstructRAG2025.05 | 49 | — | — | 42 | — | |
| AuthChainModel=Llama3-8B, Defense Framework=InstructRAG2025.05 | 49 | — | — | 44 | — | |
| InstructRAGTop-k=10, Pos=Pos 10, Generator LLM=Grok-4-fast2026.05 | 48.6 | — | — | 5.4 | — | |
| AuthChainModel=Llama3-70B, Defense Framework=InstructRAG2025.05 | 47 | — | — | 45 | — | |
| HRAGModel=GPT-3.5, Defense Framework=InstructRAG2025.05 | 45 | — | — | 50 | — | |
| AuthChainModel=GPT-3.5, Defense Framework=AstuteRAG2025.05 | 44 | — | — | 47 | — |