General LLM Capability Evaluation on MMLU, AlpacaEval, GSM8K, MATH, HumanEval
71.86MMLUBFPO
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| BFPOBackbone=Qwen2.5-7B, Temperature=1.02026.06 | 71.86 | 94.91 | 70.66 | 25.6 | 82.93 | 69.19 | |
| BFPOBase Model=Qwen2.5-7B-Instruct, Temperature=02026.06 | 71.86 | 94.91 | 71.49 | 25.6 | 83.54 | 69.48 | |
| NSPOBackbone=Qwen2.5-7B, Temperature=1.02026.06 | 71.76 | 95.28 | 69.37 | 26.6 | 79.88 | 68.58 | |
| NSPOBase Model=Qwen2.5-7B-Instruct, Temperature=02026.06 | 71.76 | 95.28 | 70.51 | 23.8 | 82.93 | 68.86 | |
| Qwen2.5-7B-InstructBackbone=Qwen2.5-7B, Temperature=1.02026.06 | 71.7 | 95.96 | 70.12 | 25.6 | 79.26 | 68.53 | |
| Qwen2.5-7B-InstructTemperature=02026.06 | 71.7 | 95.96 | 70.35 | 24.8 | 81.7 | 68.9 | |
| SafeSteerBackbone=Qwen2.5-7B, Temperature=1.02026.06 | 71.63 | 93.98 | 70.58 | 25.4 | 76.83 | 67.68 | |
| SafeSteerBase Model=Qwen2.5-7B-Instruct, Temperature=02026.06 | 71.63 | 93.98 | 72.71 | 25.2 | 82.32 | 69.17 | |
| DPO-MixBackbone=Qwen2.5-7B, Temperature=1.02026.06 | 71.61 | 54.04 | 75.51 | 26.4 | 75.61 | 60.63 | |
| DPO-MixBase Model=Qwen2.5-7B-Instruct, Temperature=02026.06 | 71.61 | 59.13 | 77.1 | 25.6 | 73.78 | 61.44 | |
| SafeSteerBackbone=Qwen2-1.5B, Temperature=1.02026.06 | 70.67 | 94.78 | 72.78 | 27.8 | 89.63 | 71.13 | |
| SafeSteerBase Model=Qwen3-4B-Instruct, Temperature=02026.06 | 70.67 | 94.78 | 73.47 | 27.2 | 90.85 | 71.39 | |
| NSPOBackbone=Qwen2-1.5B, Temperature=1.02026.06 | 70.65 | 95.78 | 73.69 | 26.4 | 87.8 | 70.86 | |
| NSPOBase Model=Qwen3-4B-Instruct, Temperature=02026.06 | 70.65 | 95.78 | 73.24 | 26.6 | 90.85 | 71.42 | |
| Qwen2-1.5B-InstructBackbone=Qwen2-1.5B, Temperature=1.02026.06 | 70.61 | 96.02 | 74.07 | 26.2 | 89.63 | 71.31 | |
| DPO-MixBackbone=Qwen2-1.5B, Temperature=1.02026.06 | 70.61 | 58.49 | 74.22 | 27.2 | 89.63 | 64.03 | |
| Qwen3-4B-InstructTemperature=02026.06 | 70.61 | 96.02 | 73.08 | 26.6 | 92.07 | 71.68 | |
| DPO-MixBase Model=Qwen3-4B-Instruct, Temperature=02026.06 | 70.61 | 68.76 | 75.89 | 29.4 | 90.85 | 67.1 | |
| BFPOBackbone=Qwen2-1.5B, Temperature=1.02026.06 | 70.6 | 95.28 | 73.01 | 26 | 90.85 | 71.15 | |
| BFPOBase Model=Qwen3-4B-Instruct, Temperature=02026.06 | 70.6 | 95.28 | 73.31 | 28 | 90.24 | 71.49 | |
| MoCANBackbone=Qwen2.5-7B, Temperature=1.02026.06 | 70.18 | 94.16 | 68.68 | 25.4 | 82.92 | 68.27 | |
| MoCANBase Model=Qwen2.5-7B-Instruct, Temperature=02026.06 | 70.18 | 94.16 | 68.91 | 24.2 | 84.76 | 68.44 | |
| W-DOORBackbone=Qwen2.5-7B, Temperature=1.02026.06 | 68 | 84.76 | 61.71 | 26.2 | 30.49 | 54.23 | |
| W-DOORBase Model=Qwen2.5-7B-Instruct, Temperature=02026.06 | 68 | 84.76 | 63.84 | 27.6 | 26.83 | 54.21 | |
| W-DOORBackbone=Qwen2-1.5B, Temperature=1.02026.06 | 65.26 | 90.99 | 40.18 | 14.2 | 24.39 | 47 | |
| W-DOORBase Model=Qwen3-4B-Instruct, Temperature=02026.06 | 65.26 | 90.99 | 42.29 | 14 | 27.44 | 48 | |
| Llama-3-8B-InstructBase Model=Llama-3-8B-Instruct, Temperature=02026.06 | 64.72 | 96.15 | 66.34 | 30 | 60.98 | 63.64 | |
| Llama-3-8B-InstructBase Model=Llama-3-8B, Temperature=1.02026.06 | 64.72 | 96.15 | 61.52 | 24.4 | 54.27 | 60.21 | |
| BFPOBase Model=Llama-3-8B-Instruct, Temperature=02026.06 | 64.66 | 95.9 | 63.76 | 29.8 | 60.98 | 63.02 | |
| BFPOBase Model=Llama-3-8B, Temperature=1.02026.06 | 64.66 | 95.9 | 60.73 | 27.4 | 56.71 | 61.08 | |
| SafeSteerBase Model=Llama-3-8B-Instruct, Temperature=02026.06 | 64.63 | 95.27 | 65.04 | 30 | 60.98 | 63.18 | |
| SafeSteerBase Model=Llama-3-8B, Temperature=1.02026.06 | 64.63 | 95.27 | 63.68 | 23 | 53.66 | 60.05 | |
| NSPOBase Model=Llama-3-8B-Instruct, Temperature=02026.06 | 64.58 | 95.84 | 64.82 | 30.4 | 59.76 | 63.08 | |
| NSPOBase Model=Llama-3-8B, Temperature=1.02026.06 | 64.58 | 95.84 | 61.94 | 12.8 | 57.93 | 58.62 | |
| DPO-MixBase Model=Llama-3-8B-Instruct, Temperature=0, Mixing Ratio=50/502026.06 | 64.4 | 79.25 | 73.16 | 28.8 | 60.37 | 61.2 | |
| DPO-MixBase Model=Llama-3-8B, Temperature=1.02026.06 | 64.4 | 76.52 | 71.49 | 24.6 | 59.76 | 59.35 | |
| MoCANBase Model=Llama-3-8B-Instruct, Temperature=02026.06 | 63.66 | 94.35 | 71.11 | 26.2 | 57.93 | 62.65 | |
| MoCANBackbone=Qwen2-1.5B, Temperature=1.02026.06 | 62.98 | 95.14 | 72.63 | 27.2 | 89.02 | 69.39 | |
| MoCANBase Model=Qwen3-4B-Instruct, Temperature=02026.06 | 62.98 | 95.14 | 72.18 | 26.2 | 90.24 | 69.35 | |
| Llama-3.2-3B-InstructBase Model=Llama-3.2-3B-Instruct, Temperature=02026.06 | 62.19 | 92.8 | 72.78 | 37.6 | 53.67 | 63.61 | |
| Llama-3.2-3B-InstructBase Model=Llama-3.2-3B, Temperature=1.02026.06 | 62.19 | 92.8 | 63.08 | 35.4 | 46.95 | 60.08 | |
| BFPOBase Model=Llama-3.2-3B-Instruct, Temperature=02026.06 | 62.18 | 90.99 | 72.33 | 37.4 | 52.44 | 63.07 | |
| NSPOBase Model=Llama-3.2-3B-Instruct, Temperature=02026.06 | 62.18 | 91.49 | 71.95 | 38 | 52.44 | 63.21 | |
| BFPOBase Model=Llama-3.2-3B, Temperature=1.02026.06 | 62.18 | 90.99 | 60.73 | 34.8 | 53.05 | 60.35 | |
| NSPOBase Model=Llama-3.2-3B, Temperature=1.02026.06 | 62.18 | 91.49 | 63.15 | 32.2 | 49.39 | 59.68 | |
| DPO-MixBase Model=Llama-3.2-3B-Instruct, Temperature=0, Mixing Ratio=50/502026.06 | 62.02 | 79.75 | 74.68 | 37.4 | 54.27 | 61.62 | |
| DPO-MixBase Model=Llama-3.2-3B, Temperature=1.02026.06 | 62.02 | 63.69 | 62.55 | 32.6 | 48.78 | 53.93 | |
| SafeSteerBase Model=Llama-3.2-3B-Instruct, Temperature=02026.06 | 61.64 | 90.8 | 72.02 | 36.8 | 53.04 | 62.86 | |
| SafeSteerBase Model=Llama-3.2-3B, Temperature=1.02026.06 | 61.16 | 90.8 | 62.16 | 23.8 | 40.24 | 55.63 | |
| W-DOORBase Model=Llama-3-8B-Instruct, Temperature=02026.06 | 59.81 | 84.16 | 1.44 | 13.2 | 30.49 | 37.82 | |
| W-DOORBase Model=Llama-3-8B, Temperature=1.02026.06 | 59.81 | 84.16 | 1.44 | 13.2 | 32.32 | 38.19 | |
| MoCANBase Model=Llama-3.2-3B-Instruct, Temperature=02026.06 | 59.76 | 87.44 | 54.06 | 33.4 | 54.27 | 57.79 | |
| W-DOORBase Model=Llama-3.2-3B-Instruct, Temperature=02026.06 | 55.83 | 79.32 | 3.56 | 0.4 | 31.1 | 34.04 | |
| W-DOORBase Model=Llama-3.2-3B, Temperature=1.02026.06 | 55.83 | 79.32 | 2.43 | 14 | 21.34 | 34.58 | |
| MoCANBase Model=Llama-3-8B, Temperature=1.02026.06 | 54.54 | 94.35 | 62.7 | 22.6 | 56.1 | 58.06 | |
| MoCANBase Model=Llama-3.2-3B, Temperature=1.02026.06 | 50.92 | 87.44 | 41.69 | 22.2 | 40.85 | 48.62 |