Safety Alignment and Instruction Following on REDORCA
91.08HHDualSelect
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| DualSelectModel=Llama-3-8B2026.06 | 91.08 | 84.88 | 87.98 | 64.92 | |
| DualSelectModel=Qwen3-4B2026.06 | 90.58 | 84.86 | 87.72 | 64.38 | |
| SQSDModel=Qwen3-4B2026.06 | 85.86 | 78.94 | 82.4 | 64.72 | |
| DualSelectModel=Gemma-3-1B2026.06 | 84.38 | 82.71 | 83.55 | 64.28 | |
| SQSDModel=Llama-3-8B2026.06 | 84.12 | 80.24 | 82.18 | 63.76 | |
| TOSSModel=Qwen3-4B2026.06 | 79.52 | 85.24 | 82.38 | 62.48 | |
| TOSSModel=Llama-3-8B2026.06 | 78.84 | 85.36 | 82.1 | 62.72 | |
| TOSSModel=Gemma-3-1B2026.06 | 78.42 | 76.18 | 77.3 | 61.64 | |
| SQSDModel=Gemma-3-1B2026.06 | 73.86 | 83.04 | 78.45 | 63.06 | |
| LARFModel=Qwen3-4B2026.06 | 66.43 | 69.57 | 68 | 60.74 | |
| LARFModel=Llama-3-8B2026.06 | 65.68 | 68.89 | 67.29 | 60.12 | |
| LARFModel=Gemma-3-1B2026.06 | 64.69 | 68.52 | 66.61 | 58.31 | |
| SEALModel=Qwen3-4B2026.06 | 63.88 | 66.39 | 65.14 | 59.46 | |
| SEALModel=Llama-3-8B2026.06 | 62.47 | 65.61 | 64.04 | 58.64 | |
| SEALModel=Gemma-3-1B2026.06 | 61.21 | 66.44 | 63.83 | 57.86 | |
| RandomModel=Qwen3-4B2026.06 | 56.06 | 58.05 | 57.06 | 50.68 | |
| RandomModel=Llama-3-8B2026.06 | 55.68 | 57.49 | 56.59 | 50.38 | |
| SafeInstrModel=Qwen3-4B2026.06 | 55.36 | 60.51 | 57.94 | 53.24 | |
| RandomModel=Gemma-3-1B2026.06 | 55.18 | 56.34 | 55.76 | 50.42 | |
| SafeInstrModel=Gemma-3-1B2026.06 | 54.93 | 60.58 | 57.76 | 51.42 | |
| SafeInstrModel=Llama-3-8B2026.06 | 54.12 | 59.08 | 56.6 | 52.88 | |
| Standard SFTModel=Gemma-3-1B2026.06 | 50 | 50 | 50 | 50 | |
| Standard SFTModel=Qwen3-4B2026.06 | 50 | 50 | 50 | 50 | |
| Standard SFTModel=Llama-3-8B2026.06 | 50 | 50 | 50 | 50 |