FLORES
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
FLORES xx→en dupe noise v1.0 (test)
-5.5Score (de->en)
4
FLORES xx→en, drop noise v1.0 (test)
-14.4DE->EN Score
4
FLORES xx→en, swap noise v1.0 (test)
-51.4Score (de->en)
4
FLORES eng-mri
24.28BLEU
3
FLORES mri-eng
28.14BLEU
3
FLORES-200 72 languages
16.21o-xsim++
3
FLORES-200 72 languages
0.31o-xsim
3
FLORES-200 ∩ CLD3 95 languages (test)
98.9F1 Score
3
FLORES-200
100Accuracy
3
FLORES-101 source: ta (devtest)
18.2spBLEU (ta->ar)
3
FLORES-101 source: en (devtest)
32spBLEU (ar)
3
FLORES-101 source: ar (devtest)
35.5spBLEU (fr)
3
FLORES X-En
92# Improved
2
FLORES-200 pol_Latn (devtest)
23.9Tie Rate (%)
2
FLORES-200 tur_Latn (devtest)
16Tie Rate (%)
2
FLORES-200 (dev)
28.48BLEU
2
FLORES COMET (test)
0.677Kendall Tau
2
FLORES BLEU (test)
0.803Kendall Tau (τ)
2
FLORES-200 ∩ NLLB 193 languages (test)
95.9F1 Score
2
FLORES-101 source: mr (devtest)
14.1spBLEU (ar)
2
FLORES+ (test)
1Loose Accuracy
1
FLORES-200 201 languages (test)
92.7F1
1
Flores-200 Urdu (devtest)
0.879BPC df
1
Flores-200 Basque (devtest)
1.243BPC (df)
1
Flores-200 Vietnamese (devtest)
0.794BPC (df)
1