Knowledge Editing
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
Knowledge Editing Aggregate Performance
88.57Average EM
15
Knowledge Editing Locality
100Metric T
15
Knowledge Editing (Generality)
99.38T-EM
15
Knowledge Editing Reliability
99.72Target-based EM
15
Knowledge Editing (Edit on English, Test on Chinese) 1.0 (ZH)
80.23Reliability
7
Knowledge Editing Edit on English, Test on Russian 1.0 (RU)
84.28Reliability
7
Knowledge Editing Edit on English, Test on German 1.0 (DE)
90.45Reliability
7
Knowledge Editing Edit on English, Test on Czech 1.0 (CS)
87.61Reliability
7
Knowledge Editing Edit on English, Test on English 1.0
100Reliability
7
Knowledge Editing MMLU ScienceQA GSM8K (held-out)
89.2Edit Success
3
Knowledge Editing Edit on Chinese, Test on English
45.76Reliability
3