TOFU
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
TOFU Retain95
2.6Refusal Rate
6
TOFU (Retain99)
3.2Refusal Rate
6
TOFU (GAKLR)
33Forget Rate
6
TOFU GAGDR
62Forget Rate
6
TOFU Llama3.1-8B (forget01)
99Forget Set RL Score
6
TOFU (Forget 10%)
7.456RQ
5
TOFU (Forget 5%)
7.456RQ
5
TOFU (forget05)
1.99Reconstruction Loss (l_r)
5
TOFU
98.13Accuracy
5
TOFU Llama3.1-8B (forget05)
35Forget Set Accuracy (RL)
4
TOFU
-1.3Forget Accuracy
3
TOFU Chinese 10% forget
6.7Refusal Rate (Forget)
2
TOFU Chinese (5% forget)
10Refusal Rate (Forget)
2
TOFU Chinese (1% forget split)
10Refusal Rate (Forget)
2
TOFU Italian 10% (forget)
17Refusal Rate (Forget)
2
TOFU Italian (5% forget split)
23.5Refusal Rate (Forget)
2
TOFU Italian (1% forget)
32.5Refusal Rate (Forget)
2
TOFU Spanish (10% forget split)
94Refusal Rate (Forget)
2
TOFU Spanish 5% forget split
95.5Refusal Rate (Forget)
2
TOFU Spanish 1% forget
100Refusal Rate (Forget)
2
Sequential TOFU v1.0 (Request 3)
15.5S.U.
2
Sequential TOFU v1.0 (Request 1)
12.5Sequential Utility (S.U.)
2
TOFU
0Leakage@1 (ES)
2