List your company

Knowledge Evaluation on WikiText (eval)

0.777BPBDisagreement
0.771160.810580.850.88942Jan 31, 2026

Evaluation results

MethodLinks
DisagreementProtocol=Training-time2026.010.777
CADProtocol=Inference-time2026.010.779
BaselineArchitecture=LoRA, Loss=Vanilla CE, Protocol=Training-time2026.010.784
GAME-LoRAProtocol=Training-time2026.010.786
MEProtocol=Training-time2026.010.825
ActDecProtocol=Inference-time2026.010.923