Scaling Law Fitting Accuracy on Pile 17 domains 1.0 (train test)
1.533MRE (%)Extended Linear Regression Model (Eq. 3)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Extended Linear Regression Model (Eq. 3)#Param=5K2026.06 | 1.533 | 0.034 | |
| Extended Linear Regression Model (Eq. 1)#Param=3K2026.06 | 2.064 | 0.051 | |
| Additive#Param=K(2K + 1)2026.06 | 2.209 | 0.052 | |
| BiMix#Param=2K2026.06 | 2.963 | 0.144 | |
| RegMix#Param=K^22026.06 | 6.48 | 0.136 | |
| Exponential#Param=K(K + 2)2026.06 | 6.99 | 0.059 |