Language Modelling on SCROLLS
Metric: CNLI (higher is better)
Metric: CNLI (higher is better)
| # | Model↕ | CNLI▼ | Extra Data | Paper | Date↕ | Code |
|---|---|---|---|---|---|---|
| 1 | UL2 20B | 88.7 | No | UL2: Unifying Language Learning Paradigms | 2022-05-10 | Code |
| 2 | CoLT5 XL | 88.4 | No | CoLT5: Faster Long-Range Transformers with Condi... | 2023-03-17 | - |
| 3 | LongT5 XL | 88.2 | No | LongT5: Efficient Text-To-Text Transformer for L... | 2021-12-15 | Code |
| 4 | LongT5 Large | 87.3 | No | LongT5: Efficient Text-To-Text Transformer for L... | 2021-12-15 | Code |
| 5 | BART-large SLED | 87.3 | No | Efficient Long-Text Understanding with Short-Tex... | 2022-08-01 | Code |
| 6 | BART-LS | 87.1 | No | Adapting Pretrained Text-to-Text Models for Long... | 2022-09-21 | Code |
| 7 | LongT5 Base | 85.6 | No | LongT5: Efficient Text-To-Text Transformer for L... | 2021-12-15 | Code |
| 8 | BART Base | 77.4 | No | SCROLLS: Standardized CompaRison Over Long Langu... | 2022-01-10 | Code |
| 9 | LED Base | 71.5 | No | SCROLLS: Standardized CompaRison Over Long Langu... | 2022-01-10 | Code |
| 10 | Naive | 66 | No | SCROLLS: Standardized CompaRison Over Long Langu... | 2022-01-10 | Code |