Safety Evaluation on AbstainQA (val)
26AccuracyPSO-Merging
Evaluation Results
| Method | Links | |
|---|---|---|
| PSO-MergingBackbone=Qwen2.5-1.5B2026.05 | 26 | |
| Model SwarmBackbone=Qwen2.5-1.5B2026.05 | 23 | |
| EvoGMBackbone=Qwen2.5-1.5B2026.05 | 23 | |
| Single BestBackbone=Qwen2.5-1.5B2026.05 | 21 | |
| BaseBackbone=Qwen2.5-1.5B2026.05 | 18.5 | |
| CMABackbone=Qwen2.5-1.5B2026.05 | 17 | |
| DAREBackbone=Qwen2.5-1.5B2026.05 | 9.5 | |
| TABackbone=Qwen2.5-1.5B2026.05 | 7.5 | |
| TIESBackbone=Qwen2.5-1.5B2026.05 | 7.5 | |
| Model SoupBackbone=Qwen2.5-1.5B2026.05 | 7 | |
| MTLBackbone=Qwen2.5-1.5B2026.05 | 0.5 |