TOFU
Benchmarks
TaskDatasetSOTA resultTrendResults
TOFU (5%)
86.6Forget Quality
82
TOFU Forget 10%
65Aggregation Score
81
TOFU 10% forget
-99.87Privacy Leakage
72
TOFU 1.0 (Forget10)
52Model Utility (MU)
67
TOFU Forget 5% 1.0
8.519Model Utility
60
TOFU (10%)
1Forget Quality (FQ)
60
TOFU (1%)
0Forget Quality (FQ)
59
TOFU (Forget10)
100Forget Quality (FQ)
54
TOFU Forget 1%
66Aggregation Score
54
TOFU forget05 1.0
75.85Model Utility (MU)
53
TOFU 1.0 (forget01)
80.1Average Score
53
TOFU Forget01 (1% authors)
0.99Forget Quality (Rouge-L)
48
TOFU 1.0 (Retain Set)
100ROUGE-L
48
TOFU 1.0 (Real Author)
93ROUGE-L
45
TOFU
1Forget Quality (FQ)
43
TOFU Forget10 (10% authors split)
0.99Forget Quality - Rouge-L
42
TOFU Forget05 (5% authors)
0.99Forget Quality (ROUGE-L)
42
TOFU 10% (400 samples) 1.0 (forget)
60ASR
42
TOFU World Fact 1.0
0.884ROUGE-L
34
TOFU
89Forget Quality
33
TOFU Forget05 Phi-1.5B model (5%)
51.52Model Utility (MU)
32
TOFU Avg.
14.32Log Forget Quality (-log(FQ))
30
TOFU 1% forget
2.896-log(FQ)
30
TOFU 5% forget split
13.955-log(FQ)
30
TOFU (Forget01)
-396.8fLP
30