Sequential Model Editing on Counterfact
99.88EfficacyRECT (aligned)
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| RECT (aligned)Model=LLaMA32026.05 | 99.88 | 94.34 | 81.56 | 628.75 | 33.22 | |
| PRUNE (aligned)Model=LLaMA32026.05 | 99.87 | 94.91 | 79.9 | 628.03 | 33.41 | |
| MEMIT (aligned)Model=LLaMA32026.05 | 99.85 | 95.29 | 79.98 | 626.63 | 33.31 | |
| PRUNE (aligned)Model=GPT-J2026.05 | 99.83 | 96.16 | 74.77 | 617.04 | 42.05 | |
| AlphaEdit (aligned)Model=Qwen2.52026.05 | 99.83 | 92.21 | 83.06 | 625.3 | 32.09 | |
| MEMIT (aligned)Model=GPT-J2026.05 | 99.82 | 96.51 | 74.92 | 616.39 | 41.79 | |
| MEMIT (aligned)Model=Qwen2.52026.05 | 99.82 | 92.84 | 82.64 | 624.81 | 31.74 | |
| RECT (aligned)Model=GPT-J2026.05 | 99.78 | 95.69 | 76.71 | 620.73 | 41.92 | |
| PRUNE (aligned)Model=Qwen2.52026.05 | 99.78 | 91.85 | 82.25 | 625 | 31.58 | |
| EvoEditModel=Llama-3-3B2025.10 | 99.77 | 97.27 | 71.32 | 630.71 | 34.1 | |
| AlphaEdit (aligned)Model=GPT-J2026.05 | 99.73 | 96.03 | 75.49 | 617.73 | 41.95 | |
| RECT (aligned)Model=Qwen2.52026.05 | 99.72 | 91.3 | 82.8 | 625.04 | 31.76 | |
| EvoEditModel=Llama-3-8B2025.10 | 99.67 | 94.93 | 69.99 | 623.09 | 32.64 | |
| AlphaEditModel=Qwen-7B2025.10 | 99.57 | 79.81 | 82.65 | 626.8 | 30.98 | |
| NSEModel=GPT-J2024.10 | 99.55 | 91.92 | 78.96 | 620.49 | 40.24 | |
| AlphaEdit (aligned)Model=GPT-2 XL2026.05 | 99.53 | 94.1 | 66.5 | 597.86 | 39.19 | |
| EvoEditModel=Qwen-7B2025.10 | 99.5 | 82.5 | 82.73 | 626.8 | 31.22 | |
| AlphaEditModel=Llama-3-3B2025.10 | 99.42 | 96.69 | 64.71 | 624.84 | 32.92 | |
| PRUNE (aligned)Model=GPT-2 XL2026.05 | 99.32 | 93.95 | 65.04 | 583.82 | 37.36 | |
| RECT (aligned)Model=GPT-2 XL2026.05 | 99.15 | 93.82 | 66.29 | 588.71 | 37.89 | |
| MEMIT (aligned)Model=GPT-2 XL2026.05 | 99.05 | 93.46 | 64.46 | 571.72 | 35.93 | |
| AlphaEdit (aligned)Model=LLaMA32026.05 | 98.92 | 93.93 | 68.57 | 621.85 | 32.31 | |
| AlphaEditModel=Llama-3-8B2025.10 | 98.9 | 94.22 | 67.88 | 622.49 | 32.4 | |
| MEMITModel=GPT-J2024.10 | 98.55 | 95.5 | 63.64 | 546.28 | 34.89 | |
| EvoEditBackbone=Llama-3-8B, Number of sequential edits=10,000, Batch size=1002025.10 | 98.29 | 91.21 | 63.91 | 613.88 | 33.22 | |
| NSEModel=GPT2-XL2024.10 | 96.8 | 87.72 | 72.1 | 622.85 | 40.04 | |
| NSEModel=Llama32024.10 | 96.14 | 78.42 | 87.66 | 632.72 | 30.2 | |
| GRACEModel=GPT-J2024.10 | 95.88 | 0.05 | 82.11 | 620.21 | 28.53 | |
| MEMITModel=GPT2-XL2024.10 | 94.7 | 85.82 | 60.5 | 477.26 | 22.72 | |
| GRACEModel=GPT2-XL2024.10 | 94.5 | 0.04 | 78.13 | 622.56 | 31.55 | |
| FTBackbone=Llama-3-8B, Number of sequential edits=10,000, Batch size=1002025.10 | 94.04 | 84.12 | 38.15 | 401.57 | 21.35 | |
| FT-LModel=GPT-J2024.10 | 92.15 | 72.38 | 43.35 | 297.92 | 6.65 | |
| GRACEModel=Llama32024.10 | 90.72 | 0.09 | 87.23 | 632.43 | 23.79 | |
| FT-LModel=Llama32024.10 | 83.33 | 67.79 | 46.63 | 233.72 | 8.77 | |
| FTModel=Llama-3-8B2025.10 | 83.33 | 67.79 | 46.63 | 233.72 | 8.77 | |
| MEMITModel=Llama-3-3B2025.10 | 76.03 | 77.4 | 60.7 | 576.23 | 20.76 | |
| ROMEModel=Llama-3-3B2025.10 | 69.88 | 63.79 | 48.88 | 656.96 | 1.65 | |
| ROMEModel=Qwen-7B2025.10 | 68.65 | 62.44 | 51.35 | 539.77 | 4.26 | |
| PRUNEModel=Llama-3-8B2025.10 | 68.25 | 64.75 | 49.82 | 418.03 | 5.9 | |
| AlphaEditBackbone=Llama-3-8B, Number of sequential edits=10,000, Batch size=1002025.10 | 66.78 | 58.27 | 51.79 | 489.91 | 4.59 | |
| InstructEditModel=Llama-3-8B2025.10 | 66.58 | 64.18 | 47.14 | 443.85 | 7.28 | |
| RECTModel=Llama-3-8B2025.10 | 66.05 | 63.62 | 61.41 | 526.62 | 20.54 | |
| MEMITModel=Llama32024.10 | 65.65 | 64.65 | 51.56 | 437.43 | 6.58 | |
| MEMITModel=Llama-3-8B2025.10 | 65.65 | 64.65 | 51.56 | 437.43 | 6.58 | |
| MEMITModel=Qwen-7B2025.10 | 65.65 | 64.65 | 51.56 | 437.43 | 6.58 | |
| ROMEModel=Llama32024.10 | 64.4 | 61.42 | 49.44 | 449.06 | 3.31 | |
| ROMEModel=Llama-3-8B2025.10 | 64.4 | 61.42 | 49.44 | 449.06 | 3.31 | |
| FT-LModel=GPT2-XL2024.10 | 63.55 | 42.2 | 57.06 | 519.35 | 10.56 | |
| MENDModel=Llama32024.10 | 63.24 | 61.17 | 45.37 | 372.16 | 4.21 | |
| MENDModel=Llama-3-8B2025.10 | 63.24 | 61.17 | 45.37 | 372.16 | 4.21 | |
| FT-WModel=Llama32024.10 | 61.23 | 62.4 | 47.05 | 492.34 | 3.57 | |
| SPHEREModel=Qwen2.5-7B2025.10 | 60.76 | 29.24 | 3.83 | 612.67 | 14.74 | |
| ROMEModel=GPT-J2024.10 | 57.5 | 54.2 | 52.05 | 589.28 | 3.22 | |
| ROMEModel=GPT2-XL2024.10 | 54.6 | 51.18 | 52.68 | 366.13 | 0.72 | |
| SPHEREModel=LLaMA3-8B2025.10 | 52.89 | 32.07 | 5.01 | 551.51 | 30.89 | |
| MENDModel=GPT2-XL2024.10 | 50.8 | 50.8 | 49.2 | 407.21 | 1.01 | |
| MEMITBackbone=Llama-3-8B, Number of sequential edits=10,000, Batch size=1002025.10 | 49.73 | 49.24 | 51.54 | 389.31 | 3.45 | |
| FT-WModel=GPT-J2024.10 | 48.35 | 31.42 | 68.71 | 587.2 | 29.41 | |
| ROMEBackbone=Llama-3-8B, Number of sequential edits=10,000, Batch size=1002025.10 | 47.57 | 48.45 | 52.52 | 465.02 | 1.83 | |
| MENDModel=GPT-J2024.10 | 46.15 | 46.22 | 53.9 | 242.41 | 3.94 | |
| AlphaEditModel=Qwen2.5-7B2025.10 | 43.92 | 24.37 | 2.32 | 479.83 | 4.67 | |
| FT-WModel=GPT2-XL2024.10 | 42.7 | 35.93 | 63.06 | 565.96 | 13.03 | |
| Pre-editedModel=GPT2-XL2024.10 | 22.23 | 24.34 | 78.53 | 626.64 | 31.88 | |
| Pre-editedModel=GPT-2 XL2026.05 | 22.23 | 24.34 | 78.53 | 626.64 | 31.88 | |
| Pre-editedModel=GPT-J2024.10 | 16.22 | 18.56 | 83.11 | 621.81 | 29.74 | |
| Pre-editedModel=GPT-J2026.05 | 16.22 | 18.56 | 83.11 | 621.81 | 29.74 | |
| FTModel=Qwen2.5-7B2025.10 | 15.44 | 4.63 | 1.46 | 214.26 | 3.15 | |
| PRUNEModel=Qwen2.5-7B2025.10 | 14.3 | 11.27 | 6.75 | 620.74 | 29.5 | |
| Pre-editedModel=Qwen2.52026.05 | 13.95 | 16.75 | 86.02 | 625.68 | 25.49 | |
| FTModel=LLaMA3-8B2025.10 | 8.4 | 2.54 | 0.03 | 409.8 | 19.35 | |
| Pre-editedModel=Llama32024.10 | 7.85 | 10.58 | 89.48 | 635.23 | 24.14 | |
| Pre-editedModel=LLaMA32026.05 | 7.85 | 10.58 | 89.48 | 635.23 | 24.14 | |
| GRACEBackbone=Llama-3-8B, Number of sequential edits=10,000, Batch size=1002025.10 | 7.83 | 10.21 | 89.2 | 568.49 | 3.61 | |
| AlphaEditModel=LLaMA3-8B2025.10 | 4.37 | 1.71 | 0.57 | 482.36 | 4.71 | |
| PRUNEModel=LLaMA3-8B2025.10 | 1.19 | 0.34 | 0.62 | 618.72 | 49.24 | |
| RECTModel=Qwen2.5-7B2025.10 | 0.64 | 0.19 | 0.09 | 368.46 | 1.35 | |
| RECTModel=LLaMA3-8B2025.10 | 0.57 | 0.29 | 0.1 | 438.83 | 9.4 | |
| Pre-editedModel=LLaMA3-8B2025.10 | 0.49 | 0.44 | 18.09 | 634.84 | 22.06 | |
| Pre-editedModel=Qwen2.5-7B2025.10 | 0.42 | 0.46 | 15.06 | 624.45 | 23.02 | |
| MEMITModel=LLaMA3-8B2025.10 | 0 | 0 | 0 | 318.19 | 4.19 | |
| MEMITModel=Qwen2.5-7B2025.10 | 0 | 0 | 0 | 370.84 | 3.59 |