Multi-task Model Merging Evaluation on Multi-task Category-level Evaluation Suite Qwen-3-4B
1.18Average ScoreMETIS
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| METISFramework=Many-Shot, Backbone=Qwen-3-4B2026.06 | 1.18 | 1.318 | 0.92 | 1.062 | 1.655 | |
| DAREFramework=Many-Shot, Backbone=Qwen-3-4B2026.06 | 1.154 | 1.136 | 0.886 | 1.087 | 1.638 | |
| ConsensusTAFramework=Many-Shot, Backbone=Qwen-3-4B2026.06 | 1.128 | 1.182 | 0.83 | 1.052 | 1.603 | |
| TIESFramework=Post-Hoc, Backbone=Qwen-3-4B2026.06 | 1.108 | 1.136 | 0.886 | 1.029 | 1.534 | |
| TIESFramework=Many-Shot, Backbone=Qwen-3-4B2026.06 | 1.087 | 1 | 0.818 | 1.044 | 1.569 | |
| Task ArithmeticFramework=Many-Shot, Backbone=Qwen-3-4B2026.06 | 1.084 | 0.682 | 0.977 | 1.069 | 1.638 | |
| ConsensusTAFramework=Post-Hoc, Backbone=Qwen-3-4B2026.06 | 1.067 | 1 | 0.75 | 1.04 | 1.534 | |
| Task ArithmeticFramework=Post-Hoc, Backbone=Qwen-3-4B2026.06 | 1.048 | 0.909 | 0.75 | 1.036 | 1.517 | |
| DAREFramework=Post-Hoc, Backbone=Qwen-3-4B2026.06 | 0.986 | 0.818 | 0.614 | 0.994 | 1.5 |