Scaling Law Extrapolation on Japanese, Indonesian, and Swahili 1-stage data only
0.79C (R^2)M3 Scaling Law
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| M3 Scaling Law#P=112024.10 | 0.79 | 0.7 | 0.91 | 0.63 | 0.8 | 0.48 | 0.72 | |
| Sedova#P=92024.10 | 0.71 | 0.49 | 0.86 | 0.68 | 0.72 | 0.21 | 0.61 | |
| ATLAS#P=72024.10 | 0.61 | 0.59 | 0.7 | 0.2 | 0.4 | 0.46 | 0.43 | |
| Muennighoff#P=72024.10 | 0.55 | 0.4 | 0.75 | 0.18 | 0.05 | 0.27 | 0.35 | |
| Chinchilla#P=52024.10 | 0.31 | 0.01 | 0.35 | 0.32 | 0.01 | 0.18 | 0.13 | |
| He#P=62024.10 | 0.21 | 0.19 | 0.9 | 1.05 | 0.42 | 1.24 | 0.1 |