Scaling-law extrapolation on TinyStories high-D holdout
0.053RMSE (log space)Ours (Eq. (2))
Evaluation Results
| Method | Links | |
|---|---|---|
| Ours (Eq. (2))2026.05 | 0.053 | |
| Chinchilla2026.05 | 0.057 | |
| Farseer2026.05 | 0.06 | |
| Muennighoff2026.05 | 0.095 | |
| M4 (D-axis)2026.05 | 0.223 | |
| BNSL k=22026.05 | 69.62 |