Neural Machine Translation on NMT 500k high-quality random sample
0BLEUAdaptive-SeqKD
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Adaptive-SeqKDTarget Model=Teacher (theta_T), Data Strategy=High-Quality (hq)2025.02 | 0 | 0.2 | -13.1 | -55.1 | |
| Adaptive-SeqKDTarget Model=Student (theta_S), Data Strategy=High-Quality (hq)2025.02 | -0.2 | 0.3 | -23.7 | -58.9 | |
| Adaptive-SeqKDTarget Model=Teacher (theta_T), Data Strategy=Random (ra)2025.02 | -1.2 | -0.2 | -7.5 | 7.7 | |
| Adaptive-SeqKDTarget Model=Student (theta_S), Data Strategy=Random (ra)2025.02 | -1.2 | -0.1 | -11.9 | 9.8 |