List your company

Question Answering on SQuAD v2 (Accuracy, Consistency, Overconfidence)

87.51AccuracyDGAO
56.93464.87272.8180.748May 12, 2026

Evaluation results

MethodLinks
DGAOModel=Qwen-2.5-14B2026.0587.5195.2810.42
SFTModel=Qwen-2.5-14B2026.0586.9294.911.16
DGAOModel=Qwen-2.5-3B2026.0585.6393.3510.22
SFTModel=Qwen-2.5-3B2026.0584.9792.9111.14
PAFTModel=Qwen-2.5-14B2026.0583.6895.4113.52
PAFTModel=Qwen-2.5-3B2026.0581.5993.5613.77
DGAOModel=Llama-3.2-11B2026.0577.2891.4617.71
DGAOModel=Gemma-3-4B2026.0575.8285.4914.8
SFTModel=Llama-3.2-11B2026.0575.6590.9518.61
DGAOModel=Llama-3.2-3B2026.0575.4489.617.57
SFTModel=Gemma-3-4B2026.0574.0985.3116.29
SFTModel=Llama-3.2-3B2026.0573.7388.9818.54
PAFTModel=Gemma-3-4B2026.0563.9284.5824.8
PAFTModel=Llama-3.2-11B2026.0560.8485.1228.76
PAFTModel=Llama-3.2-3B2026.0558.1182.6229.19