Knowledge Editing on replaced
100ReliabilityFT
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| FTModel=Llama-3-8B-Instruct, Framework=ChainEdit (w/ ours)2025.07 | 100 | 64.6 | 45.2 | 95.1 | 53.9 | 32.5 | |
| FTModel=Llama-3-8B-Instruct, Framework=w/o ours2025.07 | 100 | 13.9 | 40.6 | 97.5 | 69.3 | 41 | |
| LoRAModel=Llama-3-8B-Instruct, Framework=ChainEdit (w/ ours)2025.07 | 100 | 66.3 | 49.3 | 95.4 | 45.7 | 32 | |
| LoRAModel=Llama-3-8B-Instruct, Framework=w/o ours2025.07 | 100 | 17.2 | 37.1 | 98.8 | 45.2 | 28.6 | |
| FTModel=Qwen2.5-1.5B-Instruct, Framework=ChainEdit (w/ ours)2025.07 | 100 | 56.4 | 42.8 | 95.6 | 35.4 | 27.7 | |
| FTModel=Qwen2.5-1.5B-Instruct, Framework=w/o ours2025.07 | 100 | 19.4 | 40.7 | 96.5 | 42.3 | 31.8 | |
| LoRAModel=Qwen2.5-1.5B-Instruct, Framework=ChainEdit (w/ ours)2025.07 | 100 | 59.8 | 49 | 95.2 | 34.2 | 24.1 | |
| LoRAModel=Qwen2.5-1.5B-Instruct, Framework=w/o ours2025.07 | 100 | 19.8 | 35.9 | 94.1 | 31.5 | 17.1 | |
| MEMITModel=Llama-3-8B-Instruct, Framework=w/o ours2025.07 | 99.1 | 19.8 | 31.8 | 70.6 | 40.1 | 35.2 | |
| MEMIT-MergeModel=Llama-3-8B-Instruct, Framework=w/o ours2025.07 | 99.1 | 19.6 | 30.7 | 70.3 | 39.6 | 34.5 | |
| MEMITModel=Qwen2.5-1.5B-Instruct, Framework=w/o ours2025.07 | 99.1 | 24 | 24.2 | 58.5 | 38.2 | 30.7 | |
| MEMIT-MergeModel=Qwen2.5-1.5B-Instruct, Framework=w/o ours2025.07 | 99 | 22.5 | 24.4 | 60 | 39.1 | 31.5 | |
| MEMIT-MergeModel=Qwen2.5-1.5B-Instruct, Framework=ChainEdit (w/ ours)2025.07 | 98.1 | 68.1 | 31.4 | 59.6 | 37.8 | 30.7 | |
| MEMIT-MergeModel=Llama-3-8B-Instruct, Framework=ChainEdit (w/ ours)2025.07 | 96.8 | 65.8 | 38.5 | 70.4 | 37.8 | 32.2 | |
| MEMITModel=Llama-3-8B-Instruct, Framework=ChainEdit (w/ ours)2025.07 | 87.8 | 64.8 | 39 | 63.7 | 40.3 | 35.9 | |
| MEMITModel=Qwen2.5-1.5B-Instruct, Framework=ChainEdit (w/ ours)2025.07 | 87.8 | 56.8 | 25.3 | 53.5 | 41.5 | 33.8 |