List your company

Mathematical Reasoning on GSM8k (Accuracy, Consistency Rate, Overconfidence Rate)

58.19AccuracyDGAO
10.329222.754635.1847.6054May 12, 2026

Evaluation results

MethodLinks
DGAOModel=Qwen-2.5-14B2026.0558.1970.0228.27
SFTModel=Qwen-2.5-14B2026.0557.969.4428.57
PAFTModel=Qwen-2.5-14B2026.0549.4369.9136.73
DGAOModel=Qwen-2.5-3B2026.0542.9168.1128.41
SFTModel=Qwen-2.5-3B2026.0542.5367.528.52
PAFTModel=Qwen-2.5-3B2026.0533.8167.9736.99
DGAOModel=Llama-3.2-11B2026.0532.4960.3535.02
SFTModel=Llama-3.2-11B2026.0532.1458.7737.65
PAFTModel=Llama-3.2-11B2026.0523.9659.8446.51
DGAOModel=Llama-3.2-3B2026.0523.1958.4734.72
SFTModel=Llama-3.2-3B2026.0522.8556.7837.56
DGAOModel=Gemma-3-4B2026.0515.5944.4331.43
SFTModel=Gemma-3-4B2026.0515.2344.3931.78
PAFTModel=Llama-3.2-3B2026.0514.365846.87
PAFTModel=Gemma-3-4B2026.0512.1748.2738.66