Knowledge Editing on Temporal τ=10 GPT-J-6B (OOD)
100ReliabilityLOKI
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| LOKIBase Model=GPT-J-6B2026.06 | 100 | 36 | 82 | 73 | |
| WISEBase Model=GPT-J-6B2026.06 | 99 | 36 | 93 | 76 | |
| FTBase Model=GPT-J-6B2026.06 | 96 | 35 | 69 | 67 | |
| AlphaEditBase Model=GPT-J-6B2026.06 | 90 | 27 | 99 | 72 | |
| MEMITBase Model=GPT-J-6B2026.06 | 82 | 28 | 99 | 70 | |
| DEFERBase Model=GPT-J-6B2026.06 | 66 | 29 | 81 | 59 | |
| GRACEBase Model=GPT-J-6B2026.06 | 59 | 22 | 100 | 60 | |
| w/o EditingBase Model=GPT-J-6B2026.06 | 56 | 21 | — | — | |
| ROMEBase Model=GPT-J-6B2026.06 | 10 | 0 | 6 | 5 |