Model Merging on Multi-task Evaluation Suite (Instruction, Math, Multilingual, Safety)
1.015Average ScoreMETIS
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| METISBackbone=Llama-3.2-3B, Merging Framework=Many-Shot2026.06 | 1.015 | 0.917 | 0.872 | 1.018 | 1.245 | |
| ConsensusTABackbone=Llama-3.2-3B, Merging Framework=Many-Shot2026.06 | 0.945 | 0.917 | 0.872 | 0.907 | 1.163 | |
| ConsensusTABackbone=Llama-3.2-3B, Merging Framework=Post-Hoc2026.06 | 0.942 | 0.875 | 0.641 | 0.958 | 1.265 | |
| TIESBackbone=Llama-3.2-3B, Merging Framework=Many-Shot2026.06 | 0.938 | 0.792 | 0.923 | 0.924 | 1.143 | |
| DAREBackbone=Llama-3.2-3B, Merging Framework=Many-Shot2026.06 | 0.914 | 0.792 | 0.846 | 0.955 | 0.98 | |
| TIESBackbone=Llama-3.2-3B, Merging Framework=Post-Hoc2026.06 | 0.883 | 0.375 | 0.897 | 1.082 | 0.776 | |
| Task ArithmeticBackbone=Llama-3.2-3B, Merging Framework=Many-Shot2026.06 | 0.857 | 0.375 | 0.744 | 1.048 | 0.878 | |
| DAREBackbone=Llama-3.2-3B, Merging Framework=Post-Hoc2026.06 | 0.807 | 0.542 | 0.615 | 0.854 | 1.122 | |
| Task ArithmeticBackbone=Llama-3.2-3B, Merging Framework=Post-Hoc2026.06 | 0.706 | 0.583 | 0.385 | 0.715 | 1.122 |