Counterfact
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
Counterfact 20,000 sequential edits
98.5Efficacy
4
CounterFact
99.93Faithfulness (R2)
3
CounterFact 500
96.6Accuracy
3
CounterFact-style 40 q
67.5Accuracy
3
CounterFact
89.4FastText LID Accuracy
3