Safety on Safety-Tuned
3.06Safety-Tuned ScoreFuseLLM
Evaluation results
| Method | Links | |
|---|---|---|
| FuseLLMMerging Protocol=FuseLLM2026.01 | 3.06 | |
| LEDMerging Protocol=LED2026.01 | 2.93 | |
| FuseLLMModel Family=Qwen2.5-14B2026.01 | 2.53 | |
| TIESModel Family=Qwen2.5-14B2026.01 | 2.46 | |
| TIESModel Family=Qwen2.5-32B2026.01 | 2.31 | |
| Task ArithmeticMerging Protocol=Task Arithmetic2026.01 | 2.05 | |
| FuseLLMModel Family=Qwen2.5-32B2026.01 | 1.97 | |
| LEDModel Family=Qwen2.5-14B2026.01 | 1.84 | |
| ReasoningBase Model=Llama-3.1-8B2026.01 | 1.68 | |
| DeepSeek-R1-Distill-Qwen-14B (Reasoning)Model Family=Qwen2.5-14B2026.01 | 1.66 | |
| TIESMerging Protocol=TIES2026.01 | 1.62 | |
| LEDModel Family=Qwen2.5-32B2026.01 | 1.57 | |
| DeepSeek-R1-Distill-Qwen-32B (Reasoning)Model Family=Qwen2.5-32B2026.01 | 1.53 | |
| DAREMerging Protocol=DARE2026.01 | 1.51 | |
| Task ArithmeticModel Family=Qwen2.5-14B2026.01 | 1.42 | |
| LinearMerging Protocol=Linear2026.01 | 1.37 | |
| Task ArithmeticModel Family=Qwen2.5-32B2026.01 | 1.35 | |
| LinearModel Family=Qwen2.5-32B2026.01 | 1.33 | |
| DAREModel Family=Qwen2.5-32B2026.01 | 1.29 | |
| DAREModel Family=Qwen2.5-14B2026.01 | 1.29 | |
| LinearModel Family=Qwen2.5-14B2026.01 | 1.27 | |
| ReasonAnyModel Family=Qwen2.5-32B2026.01 | 1.23 | |
| Qwen2.5-32B-Instruct (Safety)Model Family=Qwen2.5-32B2026.01 | 1.2 | |
| Qwen2.5-14B-Instruct (Safety)Model Family=Qwen2.5-14B2026.01 | 1.1 | |
| ReasonAnyModel Family=Qwen2.5-14B2026.01 | 1.1 | |
| SafetyBase Model=Llama-3.1-8B2026.01 | 0.97 | |
| ReasonAnyMerging Protocol=ReasonAny2026.01 | 0.84 |