Safety Evaluation on WildGuardMix
0.8974Safety ScoreImplicitRM
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ImplicitRMPolicy model=Qwen3-Instruct-4B2026.03 | 0.8974 | 10.9 | |
| ImplicitRMPolicy model=Qwen2.5-Instruct-7B2026.03 | 0.8827 | 15.3 | |
| LAGAMPolicy model=Qwen3-Instruct-4B2026.03 | 0.8701 | 7.5 | |
| SelectMixPolicy model=Qwen3-Instruct-4B2026.03 | 0.8672 | 7.2 | |
| Causal-DRPolicy Model=LLaMA2-7B2026.03 | 0.865 | 11.3 | |
| SDR2Policy model=Qwen3-Instruct-4B2026.03 | 0.8534 | 5.4 | |
| Causal-IPSPolicy Model=LLaMA2-7B2026.03 | 0.842 | 8.3 | |
| SelectMixPolicy model=Qwen2.5-Instruct-7B2026.03 | 0.8412 | 9.9 | |
| LAGAMPolicy model=Qwen2.5-Instruct-7B2026.03 | 0.8366 | 9.3 | |
| SDRPolicy Model=LLaMA2-7B2026.03 | 0.83 | 6.8 | |
| SDR2Policy model=Qwen2.5-Instruct-7B2026.03 | 0.8182 | 6.9 | |
| ILDEPolicy Model=LLaMA2-7B2026.03 | 0.815 | 4.9 | |
| NaivePolicy model=Qwen3-Instruct-4B2026.03 | 0.8094 | — | |
| IPSPolicy Model=LLaMA2-7B2026.03 | 0.809 | 4.2 | |
| NaivePolicy Model=LLaMA2-7B2026.03 | 0.777 | — | |
| NaivePolicy model=Qwen2.5-Instruct-7B2026.03 | 0.7654 | — | |
| Causal-DRPolicy Model=Qwen2.5-7B2026.03 | 0.569 | 49.2 | |
| Causal-IPSPolicy Model=Qwen2.5-7B2026.03 | 0.551 | 44.6 | |
| SDRPolicy Model=Qwen2.5-7B2026.03 | 0.537 | 40.8 | |
| ILDEPolicy Model=Qwen2.5-7B2026.03 | 0.534 | 40 | |
| IPSPolicy Model=Qwen2.5-7B2026.03 | 0.516 | 35.4 | |
| NaivePolicy Model=Qwen2.5-7B2026.03 | 0.381 | — |