Knowledge Editing on filtered dataset original (test)
1ReliabilityChainEdit + FT
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| ChainEdit + FTModel=Llama-3-8B-Instruct, ChainEdit=w/ ours2025.07 | 1 | 0.784 | 0.442 | 0.989 | 0.475 | 0.348 | |
| FTModel=Llama-3-8B-Instruct, ChainEdit=w/o ours2025.07 | 1 | 0.139 | 0.406 | 0.975 | 0.693 | 0.41 | |
| LoRAModel=Llama-3-8B-Instruct, ChainEdit=w/o ours2025.07 | 1 | 0.172 | 0.371 | 0.988 | 0.452 | 0.286 | |
| ChainEdit + FTModel=Qwen2.5-1.5B-Instruct, ChainEdit=w/ ours2025.07 | 1 | 0.675 | 0.306 | 0.987 | 0.137 | 0.271 | |
| FTModel=Qwen2.5-1.5B-Instruct, ChainEdit=w/o ours2025.07 | 1 | 0.107 | 0.23 | 0.993 | 0.226 | 0.318 | |
| ChainEdit + LoRAModel=Qwen2.5-1.5B-Instruct, ChainEdit=w/ ours2025.07 | 1 | 0.784 | 0.454 | 0.992 | 0.19 | 0.239 | |
| LoRAModel=Qwen2.5-1.5B-Instruct, ChainEdit=w/o ours2025.07 | 1 | 0.114 | 0.231 | 0.989 | 0.148 | 0.2 | |
| ChainEdit + LoRAModel=Llama-3-8B-Instruct, ChainEdit=w/ ours2025.07 | 0.999 | 0.757 | 0.513 | 0.985 | 0.454 | 0.321 | |
| MEMITModel=Llama-3-8B-Instruct, ChainEdit=w/o ours2025.07 | 0.991 | 0.193 | 0.33 | 0.724 | 0.35 | 0.352 | |
| MEMIT-MergeModel=Llama-3-8B-Instruct, ChainEdit=w/o ours2025.07 | 0.99 | 0.195 | 0.346 | 0.73 | 0.369 | 0.358 | |
| MEMIT-MergeModel=Qwen2.5-1.5B-Instruct, ChainEdit=w/o ours2025.07 | 0.99 | 0.175 | 0.217 | 0.605 | 0.29 | 0.309 | |
| MEMITModel=Qwen2.5-1.5B-Instruct, ChainEdit=w/o ours2025.07 | 0.987 | 0.175 | 0.199 | 0.597 | 0.263 | 0.303 | |
| ChainEdit + MEMIT-MergeModel=Llama-3-8B-Instruct, ChainEdit=w/ ours2025.07 | 0.984 | 0.71 | 0.428 | 0.775 | 0.315 | 0.331 | |
| ChainEdit + MEMIT-MergeModel=Qwen2.5-1.5B-Instruct, ChainEdit=w/ ours2025.07 | 0.973 | 0.725 | 0.308 | 0.593 | 0.278 | 0.295 | |
| ChainEdit + MEMITModel=Llama-3-8B-Instruct, ChainEdit=w/ ours2025.07 | 0.88 | 0.748 | 0.453 | 0.653 | 0.359 | 0.345 | |
| ChainEdit + MEMITModel=Qwen2.5-1.5B-Instruct, ChainEdit=w/ ours2025.07 | 0.88 | 0.672 | 0.252 | 0.544 | 0.321 | 0.346 |