Scaling Law Fitting on StarCoder
0.98Performance (1e-5)Perturbation-aware Baseline (Exponential)
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Perturbation-aware Baseline (Exponential)Formula=a/N^α + b/D^β + c + d(D^β'/N^α')e^(γX)2026.05 | 0.98 | 0.974 | 0.961 | 0.913 | 0.384 | 0.842 | 0.26 | |
| Shannon Scaling LawFormula=aN^α log2(1 + bD^β / (c(DN)^γ + dD^δ + e))2026.05 | 0.977 | 0.976 | 0.973 | 0.901 | 0.858 | 0.937 | 0.05 | |
| Asymmetric LawFormula=aN^α / (D^β + bD^β'N^α' + c)2026.05 | 0.97 | 0.963 | 0.941 | 0.872 | 0.832 | 0.916 | 0.06 | |
| Symmetric LawFormula=aN^α / (D^β + bD^βN^α + c)2026.05 | 0.963 | 0.95 | 0.904 | 0.677 | 0.641 | 0.827 | 0.16 | |
| OpenAI LawFormula=[a/N^α + b/D^β]^-12026.05 | 0.956 | 0.942 | 0.881 | 0.507 | -8.336 | -1.01 | 4.1 | |
| Perturbation-aware Baseline (Power)Formula=a/N^α + b/D^β + c + d(D^β'/N^α')X^γ2026.05 | 0.929 | 0.899 | 0.844 | 0.605 | 0.384 | 0.732 | 0.23 | |
| Chinchilla LawFormula=a/N^α + b/D^β + c2026.05 | 0.928 | 0.903 | 0.832 | 0.507 | 0.002 | 0.634 | 0.39 |