DiacNet
Restoring the marks that make Nigerian-language text readable, searchable, and machine-usable. Two parallel architectures carry the line today: diacnet-1.1, a ByT5 sequence-to-sequence restorer that can also fix typos along the way, and diactag-1.0, a character tagger that copies your base letters and only predicts the marks, so it cannot corrupt the text it edits.
DiacNetLABSdiacnet-yor-viterbi
Yorùbá diacritic and tone restoration powered by Viterbi decoding, making text readable, searchable, and machine-usable.
DiacNetLABSdiacnet-yor-db
Yorùbá diacritic and tone restoration powered by k-nearest neighbours, making text readable, searchable, and machine-usable.
DiacNetLABSdiacnet-ig
Igbo diacritic and tone restoration powered by k-nearest neighbours, making text readable, searchable, and machine-usable.
DiacNetLABSdiacnet-yor
Yorùbá diacritic and tone restoration powered by BiLSTM networks, making text readable, searchable, and machine-usable.
DiacNetLABSdiacnet-yor-x
Yorùbá diacritic and tone restoration powered by transformer encoders, making text readable, searchable, and machine-usable.
DiacNetSTABLEdiacnet-1.0
Restores diacritics and tone marks across 10 languages — Yorùbá, Igbo, Hausa, Vietnamese, Polish and more. One byte-level ByT5 model, no per-language vocabulary.
DiacNetSTABLEdiacnet-1.1
Restores diacritics and tone marks across 10 languages — Yorùbá, Igbo, Hausa, Vietnamese, Polish and more. One byte-level ByT5 model, no per-language vocabulary.
DiacNetSTABLEdiactag-1.0
Restores diacritics and tone marks across 10 languages by classifying each character rather than generating text, so the output is your input with marks added and nothing else. 37.6M parameters, CPU-native.
Explore other families
More of the stack, organised the same way.