Mathematical Reasoning on GSM8K (ECR, GSC, APSS)
96.42ECRTopK
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| TopKalpha (α)=0.05, sample candidates per prompt (N)=5, feature map / quadratic basis (Φ(X))=[1, T(X), T(X)^2], verifier loss type (T(X))=mean verifier loss2026.03 | 96.42 | 86.52 | 1 | |
| CFC-P-Falpha (α)=0.05, sample candidates per prompt (N)=5, feature map / quadratic basis (Φ(X))=[1, T(X), T(X)^2], verifier loss type (T(X))=mean verifier loss2026.03 | 95.24 | 88.79 | 4.59 | |
| ICPalpha (α)=0.05, sample candidates per prompt (N)=5, feature map / quadratic basis (Φ(X))=[1, T(X), T(X)^2], verifier loss type (T(X))=mean verifier loss2026.03 | 95.09 | 79.85 | 4.73 | |
| Learnt CPalpha (α)=0.05, sample candidates per prompt (N)=5, feature map / quadratic basis (Φ(X))=[1, T(X), T(X)^2], verifier loss type (T(X))=mean verifier loss2026.03 | 94.91 | 88.48 | 4.01 | |
| CFCalpha (α)=0.05, sample candidates per prompt (N)=5, feature map / quadratic basis (Φ(X))=[1, T(X), T(X)^2], verifier loss type (T(X))=mean verifier loss2026.03 | 94.82 | 88.48 | 2.35 |