Long-context language understanding on LongBench (Specific Metric Selection)
28.17QasperDream Instruct
Evaluation Results
| Method | Links | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Dream Instructcaching=None2025.05 | 28.17 | 36.23 | 27.65 | 32.43 | 11.83 | 5.04 | 14.29 | 5.95 | 73 | 89.25 | 37.84 | 16.92 | 38.91 | 45.08 | 33.04 | |
| Dream Instruct + dLLM-Cachecaching=dLLM-Cache2025.05 | 26.55 | 39.86 | 27.66 | 32.09 | 11.12 | 4.4 | 13.89 | 5.51 | 73.5 | 89.59 | 36.07 | 12.05 | 39.88 | 45.57 | 32.7 | |
| LLaDA Instructcaching=None2025.05 | 16.96 | 31.31 | 14.68 | 17.6 | 11.48 | 29.24 | 21.93 | 27.58 | 65.2 | 47.98 | 40.51 | 98.17 | 65.69 | 59.57 | 39.14 | |
| LLaDA Instruct + dLLM-Cachecaching=dLLM-Cache2025.05 | 15.26 | 29.62 | 13.87 | 17.17 | 10.44 | 29.75 | 22.06 | 26.68 | 66 | 44.94 | 41.86 | 97.44 | 66.07 | 59.34 | 38.61 |