Scaling-law extrapolation on Muennighoff grid (high-D holdout)
0.044RMSE (log space)Ours (Eq. (2))
Evaluation results
| Method | Links | |
|---|---|---|
| Ours (Eq. (2))2026.05 | 0.044 | |
| Muennighoff2026.05 | 0.079 | |
| M4 (D-axis)2026.05 | 0.101 | |
| BNSL k=22026.05 | 0.103 | |
| Chinchilla2026.05 | 0.112 | |
| Farseer2026.05 | 0.113 |