Logical Reasoning on LogiQA CN (Accuracy)
34.92Accuracy (LogiQA CN)ConSA (head-wise, single-layer)
Evaluation results
| Method | Links | |
|---|---|---|
| ConSA (head-wise, single-layer)Target Sparsity (ρ)=0.50, Granularity=head-wise, Constraint Scope=single-layer, Model=1.7B2026.06 | 34.92 | |
| Dense FATarget Sparsity (ρ)=0, Model=1.7B2026.06 | 33.23 | |
| ConSA (head-wise, all-layers)Target Sparsity (ρ)=0.50, Granularity=head-wise, Constraint Scope=all-layers, Model=1.7B2026.06 | 33.08 | |
| Rule (head-wise)Target Sparsity (ρ)=0.50, Granularity=head-wise, Model=1.7B2026.06 | 32.46 | |
| ConSA (layer-wise)Target Sparsity (ρ)=0.50, Granularity=layer-wise, Model=1.7B2026.06 | 31.54 | |
| Rule (layer-wise)Target Sparsity (ρ)=0.50, Granularity=layer-wise, Model=1.7B2026.06 | 31.02 |