Value Alignment on WildGuard (test)
29.69ScoreNo Guidance
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| No GuidanceBackbone=LLAMA-3.2-3B2026.05 | 29.69 | — | |
| No GuidanceBackbone=QWEN2-1.5B2026.05 | 26.76 | — | |
| No GuidanceBackbone=GPT-22026.05 | 24.43 | — | |
| No GuidanceBackbone=MISTRAL-7B2026.05 | 23.81 | — | |
| System PromptBackbone=GPT-22026.05 | 20.82 | 14.8 | |
| ITIBackbone=QWEN2-1.5B2026.05 | 15.23 | 43.1 | |
| System PromptBackbone=LLAMA-3.2-3B2026.05 | 13.73 | 53.8 | |
| ITIBackbone=GPT-22026.05 | 13.67 | 44 | |
| ITIBackbone=LLAMA-3.2-3B2026.05 | 12.97 | 56.3 | |
| RE-ControlBackbone=GPT-22026.05 | 12.93 | 47.1 | |
| System PromptBackbone=MISTRAL-7B2026.05 | 12.93 | 45.7 | |
| System PromptBackbone=QWEN2-1.5B2026.05 | 12.8 | 52.2 | |
| RE-ControlBackbone=LLAMA-3.2-3B2026.05 | 12.22 | 58.8 | |
| RE-ControlBackbone=QWEN2-1.5B2026.05 | 10.12 | 62.2 | |
| ITIBackbone=MISTRAL-7B2026.05 | 9.87 | 58.5 | |
| RE-ControlBackbone=MISTRAL-7B2026.05 | 9.57 | 59.8 | |
| DPO (LoRA)Backbone=MISTRAL-7B2026.05 | 9.15 | 61.6 | |
| DPO (LoRA)Backbone=QWEN2-1.5B2026.05 | 8.71 | 67.5 | |
| DPO (LoRA)Backbone=GPT-22026.05 | 8.44 | 65.5 | |
| DPO (LoRA)Backbone=LLAMA-3.2-3B2026.05 | 8.28 | 72.1 | |
| SVGTBackbone=LLAMA-3.2-3B2026.05 | 7.84 | 73.6 | |
| SVGTBackbone=MISTRAL-7B2026.05 | 6.57 | 72.4 | |
| SVGTBackbone=QWEN2-1.5B2026.05 | 6.18 | 76.9 | |
| SVGTBackbone=GPT-22026.05 | 2.66 | 89.1 |