TyDi QA
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
TyDi QA
69.4Accuracy
43
TyDi QA random subset of 10,000 samples
0.893Log-Probability Drop
12
TyDi QA
0.107Log-Prob Drop
12
TyDi QA No-context
42.6F1 (Arabic)
4
TyDi QA Gold Passage
73.8Arabic F1
4
TyDi QA less-mix (test)
27.4F1
3