Text Summarization on CNN/DailyMail BASE (test)
44.3ROUGE-1SCURank
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| SCURanktraining=Distilled ranking method, evaluation=mean ± std over 5 runs2026.04 | 44.3 | 20.8 | 30.9 | 51.7 | 69.9 | -2.45 | |
| BLANCtraining=Distilled ranking method, evaluation=mean ± std over 5 runs2026.04 | 43.9 | 20.6 | 30.6 | 51.4 | 69.5 | -2.45 | |
| BERTScoretraining=Distilled ranking method, evaluation=mean ± std over 5 runs2026.04 | 43.4 | 20.3 | 30.1 | 51.4 | 69 | -2.59 | |
| ROUGEtraining=Distilled ranking method, evaluation=mean ± std over 5 runs2026.04 | 43.3 | 20.1 | 30.3 | 50.9 | 69 | -2.62 | |
| GPTRanktraining=Distilled ranking method, evaluation=mean ± std over 5 runs2026.04 | 43 | 19.8 | 29.6 | 51.2 | 69.1 | -2.52 | |
| UnRanktraining=MLE-only baseline, evaluation=mean ± std over 5 runs2026.04 | 42.3 | 19.4 | 29.6 | 49.8 | 68.6 | -2.73 |