LIDstral Arabic
LIDstral Arabic classifies text across 51 language and dialect labels. Its released pipeline combines a fastText classifier, a trained neural stacker and calibrated class probabilities.
Three trained components
The fastText one-versus-all model produces independent class scores. A 64-unit LayerNorm MLP combines 144 derived features, then per-class isotonic regressors calibrate the result. The release requires all three model files, not only the fastText weights.LIDstral Arabic official model card
A closed set of labels
The CPU pipeline returns a label for each input that remains nonempty after preprocessing. Unsupported languages and mixed-language text can still receive one of the 51 labels. A high score should therefore be interpreted within the model’s supported label set.LIDstral Arabic official model card
Specifications & access
| Detail | Value and conditions |
|---|---|
| Launch date | |
| Access | Downloadable |
| Model identifiers | LIDstral-Arabic |
| Weight license | Apache 2.0LIDstral Arabic official model card |
| Output | 51 language and dialect classesLIDstral Arabic official model card |