Decoding on Common Numeracy Benchmarks
0.3RMSE ([0, 10^2])NTL-WAS
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| NTL-WAS2026.03 | 0.3 | 0.49 | 2.75 | 18.4 | 45 | |
| CONE2026.03 | 0.42 | 0.73 | 2.71 | 4.89 | 9.85 | |
| DICE2026.03 | 0.45 | 0.83 | 3.02 | 5.42 | 11.2 | |
| BioBERT2026.03 | 3.28 | 29.7 | 434 | 4.1 | 40.5 | |
| Random vectors2026.03 | 30.1 | 295.88 | 2,888.62 | 28 | 280 |