Relation Extraction on New York Times (Prec, F1)
92.4PrecisionSFT_Llama
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SFT_LlamaNote=Task-specific upper bound2026.04 | 92.4 | 91 | |
| SFT_QwenNote=Task-specific upper bound2026.04 | 90.85 | 89.5 | |
| DARE (TIES+)Transfer=Qwen to Llama2026.04 | 88.92 | 81.81 | |
| DARE (TIES+)Source Models=Qwen, Mistral, Target Model=Llama2026.04 | 88.92 | 81.81 | |
| DELLATransfer=Qwen to Llama2026.04 | 88.9 | 81.82 | |
| DELLASource Models=Qwen, Mistral, Target Model=Llama2026.04 | 88.9 | 81.82 | |
| TIES-MergingTransfer=Qwen to Llama2026.04 | 88.47 | 81.24 | |
| TIES-MergingSource Models=Qwen, Mistral, Target Model=Llama2026.04 | 88.47 | 81.24 | |
| EMR-MergingTransfer=Qwen to Llama2026.04 | 79.92 | 71.06 | |
| EMR-MergingSource Models=Qwen, Mistral, Target Model=Llama2026.04 | 79.92 | 71.06 | |
| DARE (TIES+)Target Model=Qwen2.5-7B-Instruct, Source Model (this task)=Llama-3.1-8B-Instruct2026.04 | 73.98 | 69.95 | |
| Task ArithmeticTransfer=Qwen to Llama2026.04 | 73.61 | 65.04 | |
| Task ArithmeticSource Models=Qwen, Mistral, Target Model=Llama2026.04 | 73.61 | 65.04 | |
| DELLATarget Model=Qwen2.5-7B-Instruct, Source Model (this task)=Llama-3.1-8B-Instruct2026.04 | 73.43 | 69.57 | |
| BreadcrumbsTransfer=Qwen to Llama2026.04 | 68.88 | 60.99 | |
| BreadcrumbsSource Models=Qwen, Mistral, Target Model=Llama2026.04 | 68.88 | 60.99 | |
| HeteroFusionTransfer=Qwen to Llama2026.04 | 68.46 | 62.26 | |
| HeteroFusionSource Models=Qwen, Mistral, Target Model=Llama2026.04 | 67.85 | 60.78 | |
| Llama MergeTransfer=Qwen to Llama2026.04 | 66.91 | 63.05 | |
| HeteroFusionTarget Model=Qwen2.5-7B-Instruct, Source Model (this task)=Llama-3.1-8B-Instruct2026.04 | 63.87 | 57.75 | |
| TIES-MergingTarget Model=Qwen2.5-7B-Instruct, Source Model (this task)=Llama-3.1-8B-Instruct2026.04 | 63.47 | 59.12 | |
| FuseLLMTransfer=Qwen to Llama2026.04 | 60.93 | 52.76 | |
| EMR-MergingTarget Model=Qwen2.5-7B-Instruct, Source Model (this task)=Llama-3.1-8B-Instruct2026.04 | 58.19 | 53.12 | |
| FuseLLMSource Models=Qwen, Mistral, Target Model=Llama2026.04 | 57.38 | 48.95 | |
| FuseLLMTarget Model=Qwen2.5-7B-Instruct, Source Model (this task)=Llama-3.1-8B-Instruct2026.04 | 47.12 | 38.87 | |
| Task ArithmeticTarget Model=Qwen2.5-7B-Instruct, Source Model (this task)=Llama-3.1-8B-Instruct2026.04 | 27.92 | 26.45 | |
| GACSource Models=Qwen, Mistral, Target Model=Llama2026.04 | 23.33 | 24.78 | |
| BreadcrumbsTarget Model=Qwen2.5-7B-Instruct, Source Model (this task)=Llama-3.1-8B-Instruct2026.04 | 22.97 | 22.17 | |
| UniTETarget Model=Qwen2.5-7B-Instruct, Source Model (this task)=Llama-3.1-8B-Instruct2026.04 | 18.33 | 18.15 | |
| UniTESource Models=Qwen, Mistral, Target Model=Llama2026.04 | 18.33 | 18.15 | |
| UniTETransfer=Qwen to Llama2026.04 | 14.68 | 15.87 | |
| GACTransfer=Qwen to Llama2026.04 | 14.56 | 15.51 |