Safety Alignment on ToolkitBench
75.79Safety ScoreSaFeR-ToolKit (+ SFT+DPO+GRPO) [7B]
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| SaFeR-ToolKit (+ SFT+DPO+GRPO) [7B]Model Scale=7B, Training Configuration=SFT+DPO+GRPO2026.03 | 75.79 | 64.35 | 62.1 | |
| SaFeR-ToolKit (+ SFT+DPO+GRPO) [3B]Model Scale=3B, Training Configuration=SFT+DPO+GRPO2026.03 | 67.62 | 44.94 | 48.01 | |
| Qwen2.5-VL-7BModel Scale=7B2026.03 | 52.3 | 31.87 | 6.03 | |
| Qwen2.5-VL-3BModel Scale=3B2026.03 | 30.81 | 19.04 | 1.33 |