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

PrefEval

Benchmarks

Task NameDataset NameSOTA ResultTrend
Preference-aligned Retrieval-Augmented GenerationPrefEval
Accuracy77.96
27
Question AnsweringPrefEval
Accuracy99.99
12
Personalization EvaluationPrefEval 10 injected adversarial turns
Pref Unaware Rate7.4
10
Preference evaluation via multi-choice queriesPrefEval Implicit
Accuracy69.9
8
Preference evaluation via multi-choice queriesPrefEval Explicit
Accuracy81.3
8
LLM Preference AlignmentPrefEval
AccPF68.8
7
Preference AlignmentPrefEval Implicit Preference (test)
Score19.1
6
Preference AlignmentPrefEval Implicit Choice (test)
Choice Accuracy78.6
6
Preference AlignmentPrefEval Explicit Preference (test)
LLM-Evaluated Score77.7
6
PersonalizationPrefEval (test)
Accuracy38.33
4
Showing 10 of 10 rows