List your company

Mathematical Reasoning on GSM8K (Accuracy by Sample Count)

94.39Task AccuracyDynaSteer
51.136462.365773.59584.8243May 19, 2026May 25, 2026May 31, 2026Jun 7, 2026Jun 13, 2026Jun 19, 2026Jun 26, 2026

Evaluation results

MethodLinks
DynaSteerBackbone=Qwen3-8B2026.0694.39—————
ROMEBackbone=Qwen3-8B2026.0677.94—————
MEMITBackbone=Qwen3-8B2026.0676.95—————
DELBackbone=Qwen2.5-1.5B2026.0557706767.264.961.7
NTLBackbone=Qwen2.5-1.5B2026.0555.168.965.164.664.961.7
EMOBackbone=Qwen2.5-1.5B2026.0554.569.364.166.366.459.6
DIST2LossBackbone=Qwen2.5-1.5B2026.0554.267.963.564.661.153.2
MLEBackbone=Qwen2.5-1.5B, Loss Function=Maximum Likelihood Estimation2026.055466.862.463.766.461.7
MixCEBackbone=Qwen2.5-1.5B2026.0552.866.662.565.665.657.4