Retrieval-Augmented Generation on Hotpot-ENKB5 Out-of-Domain Average
96.45Language ConsistencyQwen3-30B-A3B-Instruct-2507
Evaluation Results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Qwen3-30B-A3B-Instruct-2507Model Type=Teacher/Reference2026.07 | 96.45 | 74.46 | 79.22 | 83.37 | — | — | — | — | — | — | — | — | |
| TR-RAGBackbone=Qwen3-4B-Instruct-2507, Teacher=Qwen3-30B-A3B-Instruct-25072026.07 | 96.41 | 73.19 | 86.68 | 85.43 | — | — | — | — | — | — | — | — | |
| Naive-RLBackbone=Llama-3.2-3B-Instruct, Teacher=Llama-3.3-70B-Instruct2026.07 | 96.41 | 60.36 | 78.67 | 78.48 | 92.91 | 99.91 | 83.15 | 37.56 | 84.91 | 72.43 | 86.99 | 69.97 | |
| On-Policy DistillationBackbone=Llama-3.2-3B-Instruct, Teacher=Llama-3.3-70B-Instruct2026.07 | 96.41 | 59.56 | 75.27 | 77.08 | 92.91 | 99.91 | 79.72 | 39.39 | 82.29 | 68.24 | 84.97 | 69.18 | |
| Prompt-ControlBackbone=Llama-3.2-3B-Instruct, Teacher=Llama-3.3-70B-Instruct2026.07 | 96.4 | 45.28 | 55.77 | 65.82 | 92.9 | 99.9 | 62.57 | 27.99 | 62.09 | 49.44 | 72.52 | 59.11 | |
| Knowledge distillationBackbone=Llama-3.2-3B-Instruct, Teacher=Llama-3.3-70B-Instruct2026.07 | 96.4 | 55.77 | 71.4 | 74.52 | 92.9 | 99.9 | 76.16 | 35.37 | 79.11 | 63.69 | 82.72 | 66.32 | |
| SDFTBackbone=Llama-3.2-3B-Instruct, Teacher=Llama-3.3-70B-Instruct2026.07 | 96.4 | 59.55 | 75.76 | 77.23 | 92.9 | 99.9 | 81 | 38.09 | 82.61 | 68.9 | 85.5 | 68.96 | |
| TR-RAGBackbone=Llama-3.2-3B-Instruct, Teacher=Llama-3.3-70B-Instruct2026.07 | 96.4 | 63.47 | 80.22 | 80.03 | 92.9 | 99.9 | 83.47 | 43.47 | 85.68 | 74.76 | 87.35 | 72.71 | |
| Llama-3.2-3B-InstructBackbone=Llama-3.2-3B-Instruct2026.07 | 96.36 | 40.69 | 54.93 | 64 | 92.9 | 99.81 | 56.22 | 25.16 | 59.53 | 50.33 | 69.55 | 58.43 | |
| DITBackbone=Llama-3.2-3B-Instruct, Teacher=Llama-3.3-70B-Instruct2026.07 | 96.32 | 41.22 | 53.06 | 63.53 | 92.9 | 99.72 | 58.25 | 24.19 | 60.15 | 45.97 | 70.43 | 56.63 | |
| TranslateBackbone=Llama-3.2-3B-Instruct, Teacher=Llama-3.3-70B-Instruct2026.07 | 96.23 | 37.15 | 43.44 | 58.94 | 92.9 | 99.55 | 56.22 | 18.07 | 59.53 | 27.35 | 69.55 | 48.32 | |
| Llama-3.3-70B-InstructBackbone=Llama-3.3-70B-Instruct2026.07 | 96.11 | 70.2 | 85.68 | 84 | 92.63 | 99.59 | 90.66 | 49.73 | 87.06 | 84.3 | 90.12 | 77.87 | |
| SFTBackbone=Llama-3.2-3B-Instruct, Teacher=Llama-3.3-70B-Instruct2026.07 | 90.82 | 58.64 | 70.45 | 73.3 | 89.64 | 92 | 75.12 | 42.16 | 81.77 | 59.13 | 82.18 | 64.43 |