List your company

Numerical Reasoning on GSM8K (test)

1.13MAE (Scale 1)EMO
1.12481.15991.1951.2301May 19, 2026

Evaluation results

MethodLinks
EMOBackbone=Qwen2.5-1.5B, Loss Function=EMO2026.051.131.080.850.981.720.71
NTLBackbone=Qwen2.5-1.5B, Loss Function=NTL2026.051.161.060.841.051.530.78
DELBackbone=Qwen2.5-1.5B, Loss Function=Digit Entropy Loss2026.051.171.030.810.911.170.63
DIST2LossBackbone=Qwen2.5-1.5B, Loss Function=DIST2Loss2026.051.181.140.931.111.620.77
MixCEBackbone=Qwen2.5-1.5B, Loss Function=MixCE2026.051.251.180.911.081.720.79
MLEBackbone=Qwen2.5-1.5B, Loss Function=MLE2026.051.261.20.911.111.570.72