Alignment and Safety Evaluation on Safety
73Avg@kBase
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BaseBackbone=Qwen2.5-7B-Instruct2026.05 | 73 | 84.1 | |
| DFTBackbone=Qwen2.5-7B-Instruct2026.05 | 72.35 | 78.3 | |
| RPSFTBackbone=Qwen2.5-7B-Instruct2026.05 | 70.45 | 79.8 | |
| SFTBackbone=Qwen2.5-7B-Instruct2026.05 | 70.3 | 83.3 | |
| IWBackbone=Qwen2.5-7B-Instruct2026.05 | 69.8 | 80.8 | |
| DFTBackbone=Qwen2.5-3B-Instruct2026.05 | 66.03 | 68.2 | |
| BaseBackbone=Qwen2.5-3B-Instruct2026.05 | 65.9 | 69.1 | |
| IWBackbone=Llama-3.1-8B-Instruct2026.05 | 62.08 | 73.1 | |
| RPSFTBackbone=Qwen2.5-3B-Instruct2026.05 | 60.83 | 76.2 | |
| RPSFTBackbone=Llama-3.1-8B-Instruct2026.05 | 60.55 | 68.6 | |
| DFTBackbone=Llama-3.1-8B-Instruct2026.05 | 60.45 | 64.6 | |
| IWBackbone=Qwen2.5-3B-Instruct2026.05 | 60 | 69.7 | |
| SFTBackbone=Qwen2.5-3B-Instruct2026.05 | 59.98 | 80.2 | |
| BaseBackbone=Llama-3.1-8B-Instruct2026.05 | 59.43 | 76.6 | |
| SFTBackbone=Llama-3.1-8B-Instruct2026.05 | 57.83 | 67.8 |