Counterfact
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
Counterfact
9,387Efficacy
407
Counterfact
99.88Efficacy
81
CounterFact T = 300
99.3Efficacy
36
CounterFact batch-edit tasks
100Recall
36
CounterFact
73.06Efficacy
33
CounterFact sequential editing 10,000 Samples
99.5Efficacy Success
33
Counterfact uns
94.56Edit Success Rate
30
CounterFact
92Reliability
30
Counterfact 10,000 facts
10,000Relational Score
27
CounterFact
64.85Efficacy
26
CounterFact
79.4Reliability
26
CounterFact Factual
21,155.9Tracing Fidelity (ℓ(D, m))
24
CounterFact
96.12Efficacy
24
COUNTERFACT RS
100Efficacy
23
Counterfact (test)
89.6Probe Acc (Best Layer)
21
Counterfact Full (test)
99Rel. Accuracy
21
CounterFact
99.88Efficacy
20
Counterfact
92.96AVG Score
20
Counterfact (first 2000 edits)
99.95Accuracy
17
CounterFact larger
98.97Efficacy
14
CounterFact top
93.87Efficacy
14
COUNTERFACT
67.1Reliability
14
CounterFact T = 5000
96.6Efficacy
13
CounterFact 3,000 samples (test)
9,980Reliability
13
CounterFact
76.54Efficacy Score (%)
12