Question Answering on NQ (Acc, CF, KF, MFS)
61.2Accuracy (Acc)CR-DPO
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| CR-DPOBackbone=Llama-3.1-8B2026.05 | 61.2 | 79.4 | 73.5 | 76.4 | |
| SABERBackbone=Llama-3.2-3B2026.05 | 54.1 | 90.9 | 69.8 | 80.3 | |
| SABERBackbone=Llama-3.1-8B2026.05 | 52.1 | 73 | 65.3 | 69.2 | |
| TPCBackbone=Llama-3.1-8B2026.05 | 51.7 | 93.7 | 36.7 | 65.2 | |
| Implicit-SCRBackbone=Llama-3.1-8B2026.05 | 50.4 | 88.9 | 32.7 | 60.8 | |
| Internal-EvalBackbone=Llama-3.1-8B2026.05 | 49.6 | 66.7 | 59.2 | 62.9 | |
| TACS-LRBackbone=Llama-3.1-8B2026.05 | 47.5 | 92.1 | 20.4 | 56.2 | |
| TPCBackbone=Llama-3.2-3B2026.05 | 47.5 | 93.9 | 30.2 | 62.1 | |
| Vanilla RAGBackbone=Llama-3.1-8B2026.05 | 46.7 | 100 | 4.1 | 52 | |
| Explicit-SCRBackbone=Llama-3.1-8B2026.05 | 46.3 | 73 | 53.1 | 63 | |
| Context-EvalBackbone=Llama-3.1-8B2026.05 | 45.9 | 52.4 | 67.3 | 59.9 | |
| Vanilla RAGBackbone=Llama-3.2-3B2026.05 | 43.8 | 100 | 4.7 | 52.3 | |
| Internal-EvalBackbone=Llama-3.2-3B2026.05 | 43.8 | 56.1 | 67.4 | 61.8 | |
| Implicit-SCRBackbone=Llama-3.2-3B2026.05 | 43 | 91 | 11.4 | 51.2 | |
| Closed-BookBackbone=Llama-3.1-8B2026.05 | 41.3 | 6.3 | 100 | 53.2 | |
| TACS-LRBackbone=Llama-3.2-3B2026.05 | 40.9 | 84.8 | 13.6 | 49.2 | |
| CR-DPOBackbone=Llama-3.2-3B2026.05 | 40.1 | 56.1 | 55.8 | 55.9 | |
| Context-EvalBackbone=Llama-3.2-3B2026.05 | 39.9 | 57.6 | 44.2 | 50.9 | |
| Closed-BookBackbone=Llama-3.2-3B2026.05 | 36.4 | 7.6 | 100 | 53.8 | |
| R-TuningBackbone=Llama-3.1-8B2026.05 | 33.9 | 18.8 | 58.8 | 38.8 | |
| Explicit-SCRBackbone=Llama-3.2-3B2026.05 | 30.6 | 60.6 | 27.9 | 44.3 | |
| R-TuningBackbone=Llama-3.2-3B2026.05 | 23.1 | 10.9 | 45.7 | 28.3 |