Model Editing on ZsRE 3,000 samples
77.8Rel Score (QA Context)CrispEdit
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| CrispEditExecution=Batch, Backbone=Qwen-2.5-1.5B-Instruct2026.02 | 77.8 | 61 | 52.6 | 44 | 57.8 | 32.8 | 46.4 | 42 | 58.5 | |
| LocBF-FTBackbone=Qwen-2.5-1.5B-Instruct2026.02 | 71.4 | 52.9 | 38 | 30.6 | 59.6 | 42 | 54.6 | 44 | 54 | |
| Adam-NSCLBackbone=Qwen-2.5-1.5B-Instruct2026.02 | 62.6 | 50.5 | 21.4 | 15.3 | 59.3 | 38 | 46 | 44 | 32 | |
| CrispEditExecution=Sequential, Backbone=Qwen-2.5-1.5B-Instruct2026.02 | 55.5 | 40.7 | 77.7 | 51.6 | 59.3 | 39.5 | 46 | 42 | 59 | |
| FTBackbone=Qwen-2.5-1.5B-Instruct2026.02 | 35.4 | 29.6 | 32.2 | 25.5 | 50 | 24.8 | 49.8 | 34.5 | 35.5 | |
| UltraEditBackbone=Qwen-2.5-1.5B-Instruct2026.02 | 11.3 | 9.8 | 18.2 | 11.8 | 62.3 | 47.7 | 52.1 | 50 | 54 | |
| AlphaEditBackbone=Qwen-2.5-1.5B-Instruct2026.02 | 7.2 | 4.3 | 6.2 | 4.2 | 24.9 | 12.4 | 44.7 | 21.5 | 2 | |
| Qwen-2.5-1.5B-Instructstatus=Base Model (Unedited)2026.02 | 3.5 | 4 | 2.3 | 2 | 61.9 | 48.3 | 50.9 | 52 | 58 |