Sequential Model Editing on CounterFact (10K sequential edits, test)
94.45EfficacyRLSEdit
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| RLSEditModel=Qwen2.5-7B, Batch Size=100, Total Edits=10K2026.01 | 94.45 | 68.55 | 73.37 | 625.74 | 31.62 | |
| AlphaEditModel=Qwen2.5-7B, Batch Size=100, Total Edits=10K2026.01 | 94.1 | 70.29 | 75.29 | 623.51 | 31.37 | |
| RLSEditModel=Llama-3-8B, Batch Size=100, Total Edits=10K2026.01 | 89.94 | 72.84 | 60.56 | 615.58 | 26.27 | |
| FTModel=Llama-3-8B, Batch Size=100, Total Edits=10K2026.01 | 74.76 | 64.49 | 39.69 | 342.42 | 1.31 | |
| AlphaEditModel=Llama-3-8B, Batch Size=100, Total Edits=10K2026.01 | 66.78 | 58.27 | 51.79 | 489.91 | 4.59 | |
| FTModel=Qwen2.5-7B, Batch Size=100, Total Edits=10K2026.01 | 65.72 | 56.46 | 45.23 | 324.7 | 1.87 | |
| MEMITModel=Qwen2.5-7B, Batch Size=100, Total Edits=10K2026.01 | 53.13 | 51.39 | 51.52 | 532.38 | 1.63 | |
| MEMITModel=Llama-3-8B, Batch Size=100, Total Edits=10K2026.01 | 49.73 | 49.24 | 51.54 | 323.01 | 3.45 | |
| ROMEModel=Llama-3-8B, Batch Size=100, Total Edits=10K2026.01 | 47.57 | 48.45 | 52.52 | 465.02 | 1.83 | |
| ROMEModel=Qwen2.5-7B, Batch Size=100, Total Edits=10K2026.01 | 35.7 | 37.16 | 65.2 | 619.67 | 31.79 |