TOFU
Benchmarks
TaskDatasetSOTA resultTrendResults
TOFU (Split99)
1Forget Quality
28
TOFU Forget 5%
64Aggregation Metric
27
TOFU (Split99)
1Forget Quality
25
TOFU 5% (forget set)
-13.591FQ
25
TOFU (forget 10% data)
97.66UFE Score
24
TOFU (forget 5% data)
97.65UFE
24
TOFU (forget 1% data)
97.88UFE
24
TOFU forget10 (forget)
0.954rL
24
TOFU 1.0 (Forget05)
75.37Model Utility (MU)
24
TOFU Forget10 Phi-1.5B model
1Forget Quality (FQ)
24
TOFU Forget01 Phi-1.5B model (1%)
96Forget Quality (Rouge-L)
24
TOFU author-profile questions (held-out)
84.3Task Accuracy
22
ToFU NEK
0.76ROUGE-L
22
TOFU Llama-3.2-1B (Split99)
91.9Forget Quality
21
TOFU 20% known forget set (DFk)
47.64Recovery Rate
21
TOFU forget10 1.0
75.85MU
21
TOFU
95.1Overall Score
20
TOFU 5% forget
0.143Loss
20
TOFU
-0.1Log-scaled Forget Quality (1%)
18
TOFU Llama-3.2 Instruct (average of 1%, 5%, 10% forget sets)
-81.257FQ Score
18
TOFU 10% forget set 1.0
-238.973FQ
18
TOFU 1% (forget set)
-2.17FQ
18
TOFU
85Aggregation Metric
17
TOFU MIA Set
2.145FM
17
TOFU Neighbor Set
66.85FB Score
17