Safety Alignment on SORRY-Bench
10.22ASRLED-Merging
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LED-MergingLM=true, Math=false, Code=true2025.02 | 10.22 | — | |
| LED-MergingLM=true, Math=true, Code=true2025.02 | 10.44 | — | |
| LED-MergingLM=true, Math=true, Code=false2025.02 | 11.33 | — | |
| LED-MergingWizardLM-13B=true, WizardMath-13B=true, LLama-2-13B-Code-Alpaca=false2025.02 | 11.78 | — | |
| w/o MergingWizardLM-13B=false, WizardMath-13B=false, LLama-2-13B-Code-Alpaca=true2025.02 | 12.44 | — | |
| Model StockLM=true, Math=false, Code=true2025.02 | 12.67 | — | |
| LED-MergingWizardLM-13B=true, WizardMath-13B=false, LLama-2-13B-Code-Alpaca=true2025.02 | 15.11 | — | |
| LED-MergingWizardLM-13B=true, WizardMath-13B=true, LLama-2-13B-Code-Alpaca=true2025.02 | 17.11 | — | |
| Model StockLM=true, Math=true, Code=true2025.02 | 17.78 | — | |
| w/o MergingLM=true, Math=false, Code=false2025.02 | 18.67 | — | |
| LED-MergingModels (LM)=true, Models (Math)=true2025.02 | 24.22 | — | |
| Model StockWizardLM-13B=true, WizardMath-13B=true, LLama-2-13B-Code-Alpaca=false2025.02 | 24.67 | — | |
| Task ArithmeticLM=true, Math=true, Code=false2025.02 | 28.89 | — | |
| Task ArithmeticLM=true, Math=false, Code=true2025.02 | 31.11 | — | |
| w/o MergingWizardLM-13B=true, WizardMath-13B=false, LLama-2-13B-Code-Alpaca=false2025.02 | 32.44 | — | |
| Task ArithmeticWizardLM-13B=true, WizardMath-13B=true, LLama-2-13B-Code-Alpaca=false2025.02 | 34.67 | — | |
| Task ArithmeticWizardLM-13B=true, WizardMath-13B=false, LLama-2-13B-Code-Alpaca=true2025.02 | 35.56 | — | |
| BreadcrumbsLM=true, Math=true, Code=false2025.02 | 35.78 | — | |
| BreadcrumbsLM=true, Math=false, Code=true2025.02 | 36.89 | — | |
| Ties-MergingLM=true, Math=true, Code=false2025.02 | 37.11 | — | |
| Task ArithmeticWizardLM-13B=true, WizardMath-13B=true, LLama-2-13B-Code-Alpaca=true2025.02 | 37.78 | — | |
| BreadcrumbsWizardLM-13B=true, WizardMath-13B=true, LLama-2-13B-Code-Alpaca=false2025.02 | 38.22 | — | |
| Task ArithmeticLM=true, Math=true, Code=true2025.02 | 38.44 | — | |
| Model StockLM=true, Math=true, Code=false2025.02 | 39.55 | — | |
| Ties-MergingWizardLM-13B=true, WizardMath-13B=true, LLama-2-13B-Code-Alpaca=false2025.02 | 40 | — | |
| Ties-MergingWizardLM-13B=true, WizardMath-13B=true, LLama-2-13B-Code-Alpaca=true2025.02 | 40.22 | — | |
| BreadcrumbsWizardLM-13B=true, WizardMath-13B=false, LLama-2-13B-Code-Alpaca=true2025.02 | 40.44 | — | |
| BreadcrumbsWizardLM-13B=true, WizardMath-13B=true, LLama-2-13B-Code-Alpaca=true2025.02 | 40.44 | — | |
| BreadcrumbsLM=true, Math=true, Code=true2025.02 | 40.44 | — | |
| Ties-MergingWizardLM-13B=true, WizardMath-13B=false, LLama-2-13B-Code-Alpaca=true2025.02 | 43.56 | — | |
| Ties-MergingLM=true, Math=false, Code=true2025.02 | 46.44 | — | |
| Ties-MergingLM=true, Math=true, Code=true2025.02 | 46.44 | — | |
| w/o MergingLM=false, Math=true, Code=false2025.02 | 50.6 | — | |
| w/o MergingModels (LM)=true, Models (Math)=false2025.02 | 53.11 | — | |
| w/o MergingWizardLM-13B=false, WizardMath-13B=true, LLama-2-13B-Code-Alpaca=false2025.02 | 56.67 | — | |
| Task ArithmeticModels (LM)=true, Models (Math)=true2025.02 | 69.55 | — | |
| w/o MergingModels (LM)=false, Models (Math)=true2025.02 | 69.56 | — | |
| BreadcrumbsModels (LM)=true, Models (Math)=true2025.02 | 70.89 | — | |
| Ties-MergingModels (LM)=true, Models (Math)=true2025.02 | 79.78 | — | |
| w/o MergingLM=false, Math=false, Code=true2025.02 | 90.4 | — | |
| Qwen3-4B + SFTMask Rate=-, Training Setting=TF, Evaluation Protocol=0-shot2026.05 | — | 3.18 | |
| Qwen3-4B + SFT + WeMask(TF)Mask Rate=0.1, Training Setting=TF, Evaluation Protocol=0-shot2026.05 | — | 3.18 | |
| Qwen3-4B + SFT + WeMask(TF)Mask Rate=0.3, Training Setting=TF, Evaluation Protocol=0-shot2026.05 | — | 3.48 | |
| Qwen3-4B + SFT + WeMask(TF)Mask Rate=0.5, Training Setting=TF, Evaluation Protocol=0-shot2026.05 | — | 3.26 | |
| Qwen3-4B + SFT + WeMask(TF)Mask Rate=0.7, Training Setting=TF, Evaluation Protocol=0-shot2026.05 | — | 2.74 | |
| Qwen3-4B + SFT + WeMask(TF)Mask Rate=1.0, Training Setting=TF, Evaluation Protocol=0-shot2026.05 | — | 3.04 | |
| Qwen3-4B + WeMask(SFT)Mask Rate=0.1, Training Setting=SFT, Evaluation Protocol=0-shot2026.05 | — | 2.96 | |
| Qwen3-4B + WeMask(SFT)Mask Rate=0.3, Training Setting=SFT, Evaluation Protocol=0-shot2026.05 | — | 2.22 | |
| Qwen3-4B + WeMask(SFT)Mask Rate=0.5, Training Setting=SFT, Evaluation Protocol=0-shot2026.05 | — | 2.74 | |
| Qwen3-4B + WeMask(SFT)Mask Rate=0.7, Training Setting=SFT, Evaluation Protocol=0-shot2026.05 | — | 1.71 | |
| Qwen3-4B + WeMask(SFT)Mask Rate=1.0, Training Setting=SFT, Evaluation Protocol=0-shot2026.05 | — | 10.45 |