Knowledge Editing on ZsRE N = 10000
1ReliabilityGRACE
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| GRACEModel=Qwen2.5 7B2026.05 | 1 | 0.03 | 1 | 0.31 | |
| GRACEModel=DeepSeek 8B2026.05 | 1 | 0.02 | 1 | 0.27 | |
| HoReNModel=DeepSeek 8B2026.05 | 1 | 0.93 | 0.97 | 0.97 | |
| HoReNModel=SEA-LION 32B2026.05 | 1 | 0.9 | 0.97 | 0.96 | |
| HoReNModel=Qwen2.5 7B2026.05 | 0.99 | 0.89 | 0.99 | 0.96 | |
| HoReNModel=GPT-OSS 20B2026.05 | 0.99 | 0.66 | 0.98 | 0.86 | |
| GRACEModel=GPT-OSS 20B2026.05 | 0.6 | 0.02 | 1 | 0.23 | |
| WISEModel=SEA-LION 32B2026.05 | 0.57 | 0.54 | 1 | 0.68 | |
| GRACEModel=SEA-LION 32B2026.05 | 0.43 | 0.02 | 1 | 0.2 | |
| WISEModel=DeepSeek 8B2026.05 | 0.31 | 0.3 | 1 | 0.45 | |
| WISEModel=Qwen2.5 7B2026.05 | 0.21 | 0.2 | 0.76 | 0.32 | |
| WISEModel=GPT-OSS 20B2026.05 | 0.12 | 0.11 | 0.12 | 0.12 |