Mathematical Reasoning on GSM8K (test) (EM, ΔGSM8K)
66.26GSM8K EMEPnG
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| EPnGBackbone=OLMoE, Params. (%)=0.722026.07 | 66.26 | — | 38.43 | |
| ESFTBackbone=OLMoE, Params. (%)=5.082026.07 | 65.13 | — | 38.06 | |
| FFTBackbone=OLMoE, Params. (%)=1002026.07 | 64.97 | — | 38.54 | |
| LoRA (static)Backbone=OLMoE, Params. (%)=0.722026.07 | 64.44 | — | 36.7 | |
| Starting model2026.06 | 49.7 | — | — | |
| FProjBest Ep=1.02026.06 | 48.84 | 0.86 | — | |
| ForaBest Ep=1.02026.06 | 48.84 | 0.86 | — | |
| LwFBest Ep=2.22026.06 | 48.28 | 1.42 | — | |
| Full FTBest Ep=1.02026.06 | 47.43 | 2.27 | — | |
| EWCBest Ep=1.02026.06 | 47.26 | 2.44 | — | |
| L2-SPBest Ep=1.02026.06 | 47.13 | 2.57 | — | |
| WProjBest Ep=1.02026.06 | 47.05 | 2.65 | — |