Question Answering on OpenBookQA (Accuracy, TFLOPs)
93.6Accuracydense alignment
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| dense alignmentPair=4B → 14B, Regime=context-aware2026.06 | 93.6 | 21.54 | |
| dense alignmentPair=8B → 14B, Regime=context-aware2026.06 | 93.4 | 21.79 | |
| T2TPair=8B → 14B, Regime=context-aware2026.06 | 92.8 | 67.08 | |
| dense alignmentPair=14B → 8B, Regime=context-aware2026.06 | 92.6 | 15.64 | |
| T2TPair=4B → 14B, Regime=context-aware2026.06 | 92 | 56.24 | |
| Receiver-onlyPair=4B → 14B, Regime=context-aware2026.06 | 91.8 | 31.16 | |
| Receiver-onlyPair=8B → 14B, Regime=context-aware2026.06 | 91.8 | 31.16 | |
| Receiver-onlyPair=4B → 8B, Regime=context-aware2026.06 | 91.2 | 19.85 | |
| dense alignmentPair=4B → 8B, Regime=context-aware2026.06 | 91.2 | 12.56 | |
| Receiver-onlyPair=14B → 8B, Regime=context-aware2026.06 | 91.2 | 19.85 | |
| T2TPair=4B → 8B, Regime=context-aware2026.06 | 90.4 | 37.73 | |
| T2TPair=14B → 8B, Regime=context-aware2026.06 | 90.2 | 55.5 | |
| Ours-context-unawarePair=8B → 4B, Context Setting=context-unaware2026.06 | 90 | 6.56 | |
| dense alignmentPair=8B → 4B, Regime=context-aware2026.06 | 89.6 | 7.95 | |
| Ours-context-unawarePair=4B → 8B, Context Setting=context-unaware2026.06 | 88.8 | 9.42 | |
| dense alignmentPair=14B → 4B, Regime=context-aware2026.06 | 88.8 | 10.18 | |
| Ours-context-unawarePair=14B → 8B, Context Setting=context-unaware2026.06 | 88 | 13.28 | |
| Receiver-onlyPair=8B → 4B, Regime=context-aware2026.06 | 88 | 9.18 | |
| Receiver-onlyPair=14B → 4B, Regime=context-aware2026.06 | 88 | 9.18 | |
| C2CPair=4B → 14B, Regime=context-aware2026.06 | 87 | 6.64 | |
| T2TPair=8B → 4B, Regime=context-aware2026.06 | 87 | 33.91 | |
| C2CPair=8B → 14B, Regime=context-aware2026.06 | 87 | 7.42 | |
| C2CPair=4B → 8B, Regime=context-aware2026.06 | 85.6 | 4.5 | |
| C2CPair=14B → 8B, Regime=context-aware2026.06 | 85.6 | 7.83 | |
| Ours-context-unawarePair=14B → 4B, Context Setting=context-unaware2026.06 | 85.4 | 8.9 | |
| T2TPair=14B → 4B, Regime=context-aware2026.06 | 85 | 43.48 | |
| Ours-context-unawarePair=8B → 14B, Context Setting=context-unaware2026.06 | 83.8 | 15.85 | |
| Ours-context-unawarePair=4B → 14B, Context Setting=context-unaware2026.06 | 82.4 | 17.15 | |
| C2CPair=8B → 4B, Regime=context-aware2026.06 | 77.2 | 3.43 | |
| C2CPair=14B → 4B, Regime=context-aware2026.06 | 76.4 | 5.01 | |
| T2T-context-unawarePair=8B → 14B, Context Setting=context-unaware2026.06 | 27.4 | 72.06 | |
| T2T-context-unawarePair=4B → 14B, Context Setting=context-unaware2026.06 | 27.2 | 60.83 | |
| C2C-context-unawarePair=4B → 8B, Context Setting=context-unaware2026.06 | 27 | 14.94 | |
| T2T-context-unawarePair=14B → 8B, Context Setting=context-unaware2026.06 | 26.2 | 53.8 | |
| T2T-context-unawarePair=4B → 8B, Context Setting=context-unaware2026.06 | 24 | 37.86 | |
| T2T-context-unawarePair=8B → 4B, Context Setting=context-unaware2026.06 | 23.2 | 32.75 | |
| T2T-context-unawarePair=14B → 4B, Context Setting=context-unaware2026.06 | 20.2 | 41.95 | |
| C2C-context-unawarePair=4B → 14B, Context Setting=context-unaware2026.06 | 16.4 | 20.05 | |
| C2C-context-unawarePair=8B → 4B, Context Setting=context-unaware2026.06 | 10 | 14.76 | |
| C2C-context-unawarePair=8B → 14B, Context Setting=context-unaware2026.06 | 7.6 | 9.84 | |
| C2C-context-unawarePair=14B → 4B, Context Setting=context-unaware2026.06 | 6.6 | 17.93 | |
| C2C-context-unawarePair=14B → 8B, Context Setting=context-unaware2026.06 | 2.2 | 24.77 |