Aggregate Performance on Mit-Movie, TweetNER7, New York Times, CoNLL04, FindVehicle, and FabNER
85.67PrecisionSFT_Qwen
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SFT_QwenNote=Task-specific upper bound2026.04 | 85.67 | 84.39 | |
| SFT_LlamaNote=Task-specific upper bound2026.04 | 79.78 | 78.42 | |
| HeteroFusionTransfer=Qwen to Llama2026.04 | 73.96 | 71.38 | |
| Llama MergeTransfer=Qwen to Llama2026.04 | 69.82 | 67.6 | |
| FuseLLMTransfer=Qwen to Llama2026.04 | 66.31 | 63.15 | |
| EMR-MergingTransfer=Qwen to Llama2026.04 | 62.42 | 52.47 | |
| BreadcrumbsTransfer=Qwen to Llama2026.04 | 62.13 | 52.65 | |
| Task ArithmeticTransfer=Qwen to Llama2026.04 | 61.01 | 50.85 | |
| TIES-MergingTransfer=Qwen to Llama2026.04 | 59.5 | 47.44 | |
| DELLATransfer=Qwen to Llama2026.04 | 57.59 | 45.48 | |
| DARE (TIES+)Transfer=Qwen to Llama2026.04 | 57.16 | 45.32 | |
| GACTransfer=Qwen to Llama2026.04 | 44.06 | 41.15 | |
| UniTETransfer=Qwen to Llama2026.04 | 43.51 | 41.83 |