Question Answering on MedQA (Accuracy, TFLOPs)
72.9AccuracyT2T
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| T2TPair=14B → 8B, Regime=context-aware2026.06 | 72.9 | 55.5 | |
| T2TPair=8B → 14B, Regime=context-aware2026.06 | 72.66 | 67.08 | |
| dense alignmentPair=4B → 14B, Regime=context-aware2026.06 | 71.96 | 21.54 | |
| T2TPair=4B → 14B, Regime=context-aware2026.06 | 71.17 | 56.24 | |
| dense alignmentPair=8B → 14B, Regime=context-aware2026.06 | 70.46 | 21.79 | |
| dense alignmentPair=14B → 8B, Regime=context-aware2026.06 | 70.38 | 15.64 | |
| T2TPair=14B → 4B, Regime=context-aware2026.06 | 69.91 | 43.48 | |
| T2TPair=4B → 8B, Regime=context-aware2026.06 | 67.4 | 37.73 | |
| dense alignmentPair=4B → 8B, Regime=context-aware2026.06 | 67.24 | 12.56 | |
| T2TPair=8B → 4B, Regime=context-aware2026.06 | 66.46 | 33.91 | |
| dense alignmentPair=8B → 4B, Regime=context-aware2026.06 | 66.3 | 7.95 | |
| Receiver-onlyPair=4B → 14B, Regime=context-aware2026.06 | 64.7 | 31.16 | |
| Receiver-onlyPair=8B → 14B, Regime=context-aware2026.06 | 64.7 | 31.16 | |
| C2CPair=8B → 14B, Regime=context-aware2026.06 | 64.65 | 7.42 | |
| C2CPair=4B → 14B, Regime=context-aware2026.06 | 63 | 6.64 | |
| dense alignmentPair=14B → 4B, Regime=context-aware2026.06 | 63 | 10.18 | |
| C2CPair=14B → 8B, Regime=context-aware2026.06 | 62.69 | 7.83 | |
| C2CPair=4B → 8B, Regime=context-aware2026.06 | 56.87 | 4.5 | |
| C2CPair=8B → 4B, Regime=context-aware2026.06 | 55.07 | 3.43 | |
| C2CPair=14B → 4B, Regime=context-aware2026.06 | 53.42 | 5.01 | |
| Receiver-onlyPair=4B → 8B, Regime=context-aware2026.06 | 53 | 19.85 | |
| Receiver-onlyPair=14B → 8B, Regime=context-aware2026.06 | 53 | 19.85 | |
| Receiver-onlyPair=8B → 4B, Regime=context-aware2026.06 | 47.7 | 9.18 | |
| Receiver-onlyPair=14B → 4B, Regime=context-aware2026.06 | 47.7 | 9.18 |