Sequential Knowledge Editing on ZsRE
99.67EfficacyMEMIT (aligned)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| MEMIT (aligned)Model=GPT-J2026.05 | 99.67 | 96.63 | 28.38 | — | |
| PRUNE (aligned)Model=GPT-J2026.05 | 99.67 | 95.89 | 28.44 | — | |
| RECT (aligned)Model=GPT-J2026.05 | 99.62 | 95.89 | 28.17 | — | |
| AlphaEdit (aligned)Model=GPT-J2026.05 | 99.6 | 96 | 28.58 | — | |
| AlphaEdit (aligned)Model=Qwen2.52026.05 | 99.24 | 91.46 | 42.13 | — | |
| PRUNE (aligned)Model=Qwen2.52026.05 | 99.21 | 91.9 | 42.48 | — | |
| MEMIT (aligned)Model=Qwen2.52026.05 | 99.18 | 92.07 | 43.05 | — | |
| RECT (aligned)Model=Qwen2.52026.05 | 99.11 | 91.43 | 42.29 | — | |
| RECT (aligned)Model=LLaMA32026.05 | 95.71 | 91.58 | 32.27 | — | |
| PRUNE (aligned)Model=LLaMA32026.05 | 95.67 | 91.52 | 32.42 | — | |
| MEMIT (aligned)Model=LLaMA32026.05 | 95.55 | 91.63 | 32.42 | — | |
| RECT (aligned)Model=GPT-2 XL2026.05 | 95.46 | 87.89 | 26.33 | — | |
| DeltaEditModel=GPT2-XL2025.05 | 95.26 | 88.48 | 24.96 | — | |
| MEMIT (aligned)Model=GPT-2 XL2026.05 | 95.26 | 88.39 | 26.34 | — | |
| PRUNE (aligned)Model=GPT-2 XL2026.05 | 95.15 | 87.98 | 26.16 | — | |
| DeltaEditModel=Llama3-8B2025.05 | 94.94 | 91.19 | 30.08 | — | |
| AlphaEdit (aligned)Model=LLaMA32026.05 | 94.53 | 90.8 | 32.54 | — | |
| AlphaEditModel=Llama3-8B2025.05 | 94.49 | 91.52 | 29.77 | — | |
| AlphaEdit (aligned)Model=GPT-2 XL2026.05 | 93.42 | 84.83 | 25.67 | — | |
| AlphaEditModel=GPT2-XL2025.05 | 93.19 | 84.97 | 23.55 | — | |
| ROMEModel=GPT2-XL2025.05 | 44.81 | 41.12 | 7.52 | — | |
| RECTModel=GPT2-XL2025.05 | 39.5 | 36.13 | 15.5 | — | |
| Pre-editedModel=LLaMA32026.05 | 36.99 | 36.34 | 31.89 | — | |
| Pre-editedModel=Qwen2.52026.05 | 36.42 | 35.26 | 38.4 | — | |
| MEMITModel=GPT2-XL2025.05 | 26.8 | 25.38 | 13.51 | — | |
| Pre-editedModel=GPT-J2026.05 | 26.32 | 25.79 | 27.42 | — | |
| Pre-editedModel=GPT-2 XL2026.05 | 22.19 | 31.3 | 24.15 | — | |
| FTModel=Llama3-8B2025.05 | 6.45 | 6.01 | 24.31 | — | |
| FTModel=GPT2-XL2025.05 | 5.13 | 4.43 | 0.28 | — | |
| ROMEModel=Llama3-8B2025.05 | 2.95 | 2.82 | 1.22 | — | |
| PRUNEModel=GPT2-XL2025.05 | 2.94 | 2.77 | 3.49 | — | |
| MoEEditModel=Qwen3-30B-A3B2026.02 | 0.8447 | 0.7801 | 0.4282 | 0.6843 | |
| MoEEditModel=GPT-OSS-20B2026.02 | 0.8168 | 0.6844 | 0.3255 | 0.6089 | |
| UnKEModel=GPT-OSS-20B2026.02 | 0.4658 | 0.4399 | 0.314 | 0.4066 | |
| FT-LModel=Qwen3-30B-A3B2026.02 | 0.4419 | 0.4246 | 0.4192 | 0.4286 | |
| AdaLoRAModel=GPT-OSS-20B2026.02 | 0.4346 | 0.4296 | 0.336 | 0.4001 | |
| Pre-editedModel=Qwen3-30B-A3B2026.02 | 0.413 | 0.405 | 0.4091 | 0.409 | |
| Pre-editedModel=GPT-OSS-20B2026.02 | 0.332 | 0.3214 | 0.2802 | 0.3112 | |
| FT-LModel=GPT-OSS-20B2026.02 | 0.3275 | 0.3309 | 0.3006 | 0.3197 | |
| UnKEModel=Qwen3-30B-A3B2026.02 | 0.3143 | 0.2978 | 0.253 | 0.2884 | |
| FTModel=GPT-OSS-20B2026.02 | 0.2557 | 0.2341 | 0.1761 | 0.222 | |
| FTModel=Qwen3-30B-A3B2026.02 | 0.0644 | 0.0613 | 0.0215 | 0.0491 | |
| AdaLoRAModel=Qwen3-30B-A3B2026.02 | 0.0366 | 0.036 | 0.0468 | 0.0398 | |
| MEMITModel=Llama3-8B2025.05 | 0 | 0 | 0 | — | |
| PRUNEModel=Llama3-8B2025.05 | 0 | 0 | 0 | — | |
| RECTModel=Llama3-8B2025.05 | 0 | 0 | 0 | — |