Skip to content
Log in

Open models

Models trained by Draconitic Labs, released under the MIT licence on Hugging Face with their weights, cards and browser demos.

Draconitic Labs on Hugging Face
Models
1
Licence
MIT
Published on
Hugging Face

Language identification

Xiangliu

Names the language of a piece of text.

Hashed character n-grams into one hidden layer, trained from scratch on Tatoeba, Common Voice and SMOL. It runs in Python with NumPy or in plain JavaScript.

Model cardTry the demo
Languages
284
Model file
4.8 MB
Speed
8,922 texts a second, one CPU thread, Python
Licence
MIT
FLORES-200 test text cut to 50 characters, over the 106 languages all three models have.
ModelFileMacro F1
Xiangliu4.8 MB0.944
GlotLID v4 small35 MB0.930
fastText lid.176.bin131 MB0.837