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

Knowledge Editing

Benchmarks

Task NameDataset NameSOTA ResultTrend
Knowledge EditingKnowledge Editing Aggregate Performance
Average EM88.57
15
Knowledge EditingKnowledge Editing Locality
Metric T100
15
Knowledge EditingKnowledge Editing (Generality)
T-EM99.38
15
Knowledge EditingKnowledge Editing Reliability
Target-based EM99.72
15
Knowledge EditingKnowledge Editing (Edit on English, Test on Chinese) 1.0 (ZH)
Reliability80.23
7
Knowledge EditingKnowledge Editing Edit on English, Test on Russian 1.0 (RU)
Reliability84.28
7
Knowledge EditingKnowledge Editing Edit on English, Test on German 1.0 (DE)
Reliability90.45
7
Knowledge EditingKnowledge Editing Edit on English, Test on Czech 1.0 (CS)
Reliability87.61
7
Knowledge EditingKnowledge Editing Edit on English, Test on English 1.0
Reliability100
7
Knowledge EditingKnowledge Editing MMLU ScienceQA GSM8K (held-out)
Edit Success89.2
3
Knowledge EditingKnowledge Editing Edit on Chinese, Test on English
Reliability45.76
3
Showing 11 of 11 rows