Question Answering on ConFiQA
91.5Exact Match (EM)ProbeRAG
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ProbeRAGArchitecture=Qwen2.5-7B-Instruct2025.10 | 91.5 | 94.3 | — | |
| ProbeRAGArchitecture=LLaMA-2-7B-Chat-HF2025.10 | 81.7 | 86.1 | — | |
| MACRModel=Qwen2.5-7B2026.06 | 78 | — | 85.7 | |
| MACRModel=Llama3.1-8B2026.06 | 75 | — | 83.2 | |
| CANOEArchitecture=LLaMA-2-7B-Chat-HF2025.10 | 72.7 | 74.3 | — | |
| CANOEArchitecture=Qwen2.5-7B-Instruct2025.10 | 66.9 | 70.6 | — | |
| Context-DPOArchitecture=Qwen2.5-7B-Instruct2025.10 | 66.3 | 75 | — | |
| Context-DPOArchitecture=LLaMA-2-7B-Chat-HF2025.10 | 64.7 | 73.7 | — | |
| TruthfulRAGModel=Qwen2.5-7B2026.06 | 55.8 | — | 65.4 | |
| CK-PLUGModel=Llama3.1-8B2026.06 | 54.4 | — | 66.9 | |
| TruthfulRAGModel=Llama3.1-8B2026.06 | 53.2 | — | 62.8 | |
| InstructRAGICLModel=Llama3.1-8B2026.06 | 52.7 | — | 64 | |
| ICLModel=Llama3.1-8B2026.06 | 48.8 | — | 60.6 | |
| InstructRAGICLModel=Qwen2.5-7B2026.06 | 47.5 | — | 61.8 | |
| ICLModel=Qwen2.5-7B2026.06 | 46.6 | — | 61.6 | |
| CK-PLUGModel=Qwen2.5-7B2026.06 | 40.2 | — | 59 | |
| DirectModel=Llama3.1-8B2026.06 | 27.5 | — | 43.4 | |
| DirectModel=Qwen2.5-7B2026.06 | 19 | — | 32.2 |