TOFU
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
ToFU Neighbouring Knowledge (NEK) (one author)
78ROUGE-L
11
ToFU GEK
80Accuracy
11
TOFU (Forget05)
0.686Model Utility
11
TOFU Llama 3.1 8B (10%)
0.671Forgetting Quality
11
TOFU
0.216-ΔLr
10
TOFU Llama2-7B (unlearning)
90.8DF ROUGE
10
TOFU (forget05 retain95)
100FSR (%)
10
TOFU Llama-2-7b Llama-3.2-1B (forget)
0FKM (Knowledge Memorization Probe)
10
TOFU (test)
99.87TSR
10
TOFU
0Forget Quality
10
TOFU Llama-3-3B (unlearning)
6.042Extraction Strength
10
ToFU
0.991Probability
10
TOFU (forget01)
92Forgetting Quality (F.Q.)
10
TOFU 1.0 (Forget Set)
1ROUGE-L
10
TOFU
98.3ROUGE-L Recall
9
TOFU Forget10
8.28Relevance Score
8
TOFU (Forget05)
6.62Relevance
8
TOFU
89USR
8
TOFU Forget10
6.57FQ
7
TOFU
14.04Forgetting Score (FS)
7
TOFU 5% forget set
0FQ
7
TOFU 1.0 (Forget 10%)
100Forget Quality
7
TOFU Forget05 (retain)
55.8R-MCQA
6
TOFU Forget05 (Forget)
5.04F-Cloze
6
TOFU Retain90
0Refusal Rate
6