agbalu/Masinissa-31M
Fill-Mask • 31.1M • Updated • 102
Encoders, Embeddings & Classification
Pretrained Kabyle encoder, dense retrieval embeddings, language identification, sentiment, and tokenizers.
Note Pretrained Kabyle text encoder.
agbalu/SiMohand-278M
Sentence Similarity • 0.3B • Updated • 21Note Sentence embeddings for Kabyle semantic search and retrieval.
agbalu/Dihya-5M
Text Classification • 5.09M • Updated • 17Note Language identification across six Berber varieties.
agbalu/Masinissa-Sentiment
Text Classification • 31M • Updated • 16Note Three-class Kabyle sentiment classifier.
agbalu/Mammeri-Tok
UpdatedNote Unigram tokenizers for Kabyle.
agbalu/KabSentiment
Viewer • Updated • 15k • 95Note Balanced three-class Kabyle sentiment dataset.
agbalu/Masinissa-73M
Fill-Mask • 73.3M • Updated • 19Note Flagship 73M Kabyle encoder (20 layers). 1,953.7 Elo on KabEncBench (#1 overall, +753+ over external models); 80.59% UPOS Macro-F1 (94.08% surface).
KabEncBench Leaderboard
🏆Explore Kabyle language model benchmarks and compare performances
Note Interactive multi-task representation benchmark and Bradley–Terry Elo leaderboard.
agbalu/Dihya-22M
Text Classification • 22M • UpdatedNote Pan-Amazigh, North African Arabic dialects, and World Language Identification across 22 classes and 5 scripts.