List your company

Numerical Reasoning on GSM8K (test) (Error Magnitude Accuracy)

70Accuracy (Error <= 1)DEL
66.46467.38268.369.218May 19, 2026

Evaluation results

MethodLinks
DELBackbone=Qwen2.5-1.5B, Loss Function=Digit Entropy Loss2026.05706767.264.961.757
EMOBackbone=Qwen2.5-1.5B, Loss Function=EMO2026.0569.364.166.366.459.654.5
NTLBackbone=Qwen2.5-1.5B, Loss Function=NTL2026.0568.965.164.664.961.755.1
DIST2LossBackbone=Qwen2.5-1.5B, Loss Function=DIST2Loss2026.0567.963.564.661.153.254.2
MLEBackbone=Qwen2.5-1.5B, Loss Function=MLE2026.0566.862.463.766.461.754
MixCEBackbone=Qwen2.5-1.5B, Loss Function=MixCE2026.0566.662.565.665.657.452.8