Scaling Law Extrapolation on Japanese, Indonesian, and Swahili Monolingual Multi-epoch 1-stage (test)
0.88R^2 (C)M3+R*M(k)
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| M3+R*M(k)#P=92024.10 | 0.88 | 0.79 | 0.89 | 0.92 | 0.77 | 0.85 | |
| Sedova#P=92024.10 | 0.79 | 0.1 | 0.82 | 0.9 | 0.43 | 0.61 | |
| Muennighoff#P=72024.10 | 0.74 | 0.71 | 0.9 | 0.9 | 0.47 | 0.75 | |
| M3 Scaling Law#P=72024.10 | 0.74 | 0.71 | 0.9 | 0.9 | 0.47 | 0.75 | |
| ATLAS#P=72024.10 | 0.7 | 0.46 | 0.91 | 0.86 | 0.47 | 0.68 |