Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Winogender

Benchmarks

Task NameDataset NameSOTA ResultTrend
Attention Head Selection (Causal Tracing)WinoGender
Loss (10% Sparsity)0.0082
24
Coreference ResolutionWinogender
Accuracy64.72
15
Coreference ResolutionWinogender (test)
Accuracy80.7
11
Coreference ResolutionWinogender (WG) (test)
Accuracy80.8
11
SafetyWinogender
Normalized Log Accuracy65.3
9
Bias EvaluationWinoGender
EBS0.068
8
Commonsense ReasoningWinoGender
Accuracy0.9681
8
Multiple-choice scoringWinogender
Accuracy0.847
7
Co-reference resolutionWinoGender
Accuracy (All)77.5
4
Pronoun Coreference ResolutionWinogender (test)
Accuracy64.5
3
Showing 10 of 10 rows