Hallucination Correction on UniEdit
1Error Rate (ERR)GRACE
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| GRACEModel=LLaMA-3-8B2026.03 | 1 | 244 | 89 | |
| ELDERModel=LLaMA-3-8B2026.03 | 1 | 173 | 74 | |
| MELOModel=LLaMA-3-8B2026.03 | 1 | 153 | 68 | |
| SoLAModel=LLaMA-3-8B2026.03 | 1 | 144 | 66 | |
| GRACEModel=DeepSeek-R1-8B2026.03 | 1 | 407 | 803 | |
| ELDERModel=Qwen2-7B2026.03 | 1 | 201 | 266 | |
| MELOModel=Qwen2-7B2026.03 | 1 | 170 | 201 | |
| SoLAModel=Qwen2-7B2026.03 | 1 | 156 | 110 | |
| ELDERModel=DeepSeek-R1-8B2026.03 | 1.03 | 241 | 482 | |
| CLEARModel=Qwen2-7B2026.03 | 1.03 | 5,021 | 54.98 | |
| CLEARModel=LLaMA-3-8B2026.03 | 1.04 | 982 | 29.79 | |
| CLEARModel=DeepSeek-R1-8B2026.03 | 1.06 | 3,365 | 44.86 | |
| SoLAModel=DeepSeek-R1-8B2026.03 | 1.07 | 188 | 374 | |
| MELOModel=DeepSeek-R1-8B2026.03 | 1.09 | 204 | 407 | |
| GRACEModel=Qwen2-7B2026.03 | 2.05 | 316 | 1,065 | |
| CMRModel=DeepSeek-R1-8B2026.03 | 2.34 | 7,211 | 36.19 | |
| EWCModel=DeepSeek-R1-8B2026.03 | 2.38 | 4,451 | 42.13 | |
| EWCModel=LLaMA-3-8B2026.03 | 2.43 | 1,134 | 22.08 | |
| CMRModel=LLaMA-3-8B2026.03 | 2.79 | 2,239 | 25.24 | |
| CMRModel=Qwen2-7B2026.03 | 4.12 | 11,583 | 40.56 | |
| EWCModel=Qwen2-7B2026.03 | 4.62 | 7,648 | 48.62 | |
| SERACModel=LLaMA-3-8B2026.03 | 37.1 | 246 | 93 | |
| SERACModel=DeepSeek-R1-8B2026.03 | 46.76 | 451 | 834 | |
| SERACModel=Qwen2-7B2026.03 | 58.51 | 329 | 1,261 |