List your company

General Language Understanding on Winogrande, HellaSwag, ARC, MMLU Consolidated

71.09Average AccuracyTeacher (DeepSeek-V2-Lite)
28.616439.643250.6761.6968May 27, 2026

Evaluation results

MethodLinks
Teacher (DeepSeek-V2-Lite)2026.0571.09
DO-ACPScoring=DO-ACP, K=62026.0542.39
SFScoring=SF, K=122026.0541.16
DO-ACPScoring=DO-ACP, K=122026.0541.07
ACPScoring=ACP, K=122026.0540.93
CPScoring=CP, K=122026.0540.53
ACPScoring=ACP, K=62026.0540.37
SFScoring=SF, K=62026.0540.04
CPScoring=CP, K=62026.0538.07
Random initialization2026.0530.27
Random FFN + teacher attn2026.0530.25