List your company

Instruction Following on Pronoun Changing

99.88P. ScoreAdaSEKA
36.419252.894669.3785.8454Mar 1, 2026

Evaluation results

MethodLinks
AdaSEKAModel=Qwen3-14B2026.0399.8899.86
AdaSEKAModel=Qwen3-8B2026.0399.6899.52
PASTAModel=Qwen3-8B2026.0398.8698.72
**-markedModel=Qwen3-14B2026.0398.8698.68
SEKAModel=Qwen3-14B2026.0398.6698.54
SEKAModel=Qwen3-8B2026.0398.5698.26
OriginalModel=Qwen3-14B2026.0398.4298.22
w/o learnModel=Qwen3-8B, Configuration=Ablation: random projections2026.0398.1297.9
**-markedModel=Qwen3-8B2026.0398.197.84
OriginalModel=Qwen3-8B2026.039897.84
SEKAModel=Gemma3-12B2026.0397.797.24
AdaSEKAModel=Gemma3-12B2026.0396.8895.84
w/o learnModel=Qwen3-14B, Configuration=Ablation: random projections2026.0396.3296.36
PASTAModel=Qwen3-4B2026.0395.8294.64
**-markedModel=Gemma3-12B2026.0395.7896.42
**-markedModel=Qwen3-4B2026.0395.7693.88
SEKAModel=Qwen3-4B2026.0395.1893.26
AdaSEKAModel=Qwen3-4B2026.0394.5492.08
OriginalModel=Gemma3-12B2026.0393.9294.96
AdaSEKAModel=Gemma3-4B2026.0393.7693.58
OriginalModel=Qwen3-4B2026.0393.1490.52
SPAModel=Qwen3-14B2026.0391.692.2
PASTAModel=Qwen3-14B2026.0390.9890.94
w/o learnModel=Qwen3-4B, Configuration=Ablation: random projections2026.0390.4288.66
w/o learn&filtModel=Qwen3-14B, Configuration=Ablation: remove head filtering2026.0388.689.76
SPAModel=Gemma3-12B2026.0386.4184.99
SEKAModel=Gemma3-4B2026.0381.5381.11
w/o learn&filtModel=Qwen3-8B, Configuration=Ablation: remove head filtering2026.0380.5380.85
SPAModel=Qwen3-4B2026.0380.2778.19
SPAModel=Gemma3-4B2026.0376.0574.45
SPAModel=Qwen3-8B2026.0372.6174.95
PASTAModel=Gemma3-12B2026.0368.4768.01
PASTAModel=Gemma3-4B2026.0367.3966.43
w/o learn&filtModel=Gemma3-12B, Configuration=Ablation: remove head filtering2026.0355.5658.76
w/o learnModel=Gemma3-4B, Configuration=Ablation: random projections2026.0353.5848.82
w/o learn&filtModel=Gemma3-4B, Configuration=Ablation: remove head filtering2026.0351.7851.94
w/o learnModel=Gemma3-12B, Configuration=Ablation: random projections2026.0347.2651.24
w/o learn&filtModel=Qwen3-4B, Configuration=Ablation: remove head filtering2026.0341.9836.95
OriginalModel=Gemma3-4B2026.0341.3435.25
**-markedModel=Gemma3-4B2026.0338.8632.45