Model Editing on WikiBigEdit 3,000 samples (test)
99.9ReliabilityLocBF-FT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| LocBF-FTbackbone=LLaMA-3-8B-Instruct, pipeline=teacher-forcing2026.02 | 99.9 | 96.8 | 42.8 | |
| CrispEditbackbone=LLaMA-3-8B-Instruct, pipeline=teacher-forcing2026.02 | 99.9 | 97.1 | 44.7 | |
| CrispEdit-Seqbackbone=LLaMA-3-8B-Instruct, pipeline=teacher-forcing, editing_mode=sequential2026.02 | 99.9 | 96.7 | 39.5 | |
| FTbackbone=LLaMA-3-8B-Instruct, pipeline=teacher-forcing2026.02 | 99.8 | 97.6 | 36.3 | |
| Adam-NSCLbackbone=LLaMA-3-8B-Instruct, pipeline=teacher-forcing2026.02 | 99.7 | 97.5 | 36.4 | |
| LoRAbackbone=LLaMA-3-8B-Instruct, pipeline=teacher-forcing2026.02 | 99.3 | 82.4 | 44.4 | |
| AlphaEditbackbone=LLaMA-3-8B-Instruct, pipeline=teacher-forcing2026.02 | 95 | 89 | 42 | |
| FT Sequentialbackbone=LLaMA-3-8B-Instruct, pipeline=teacher-forcing, editing_mode=sequential2026.02 | 93.5 | 90.5 | 29 | |
| UltraEditbackbone=LLaMA-3-8B-Instruct, pipeline=teacher-forcing2026.02 | 87.4 | 84.7 | 47.8 | |
| LoRA Sequentialbackbone=LLaMA-3-8B-Instruct, pipeline=teacher-forcing, editing_mode=sequential2026.02 | 70.2 | 65.4 | 36 | |
| Llama 3 8B Instructbackbone=LLaMA-3-8B-Instruct, pipeline=teacher-forcing2026.02 | 34 | 34.8 | 32.8 | |
| MEMITbackbone=LLaMA-3-8B-Instruct, pipeline=teacher-forcing2026.02 | 0.5 | 0.5 | 0 | |
| MENDbackbone=LLaMA-3-8B-Instruct, pipeline=teacher-forcing2026.02 | 0 | 0 | 0 |