Le petit modèle sort directement du JSON, l'extraction de documents côté terminal n'a plus besoin d'écrire une petite rédaction avant de l'analyser, 450M peuvent le faire fonctionner de manière très fluide

Liquid AI open source two small multimodal models lf m2.5-vl-1.6b-extract and lf m2.5-vl-450m-extract, specifically optimized for extracting structured image data, capable of converting images directly into JSON on the device side based on field lists, eliminating the step of full text generation and parsing.
Offers two versions: 1.6b and 450m, following the lf m open license v1.0.
Official evaluations show excellent performance in document scanning, in-vehicle cabin understanding, and industrial inspection scenarios;
In benchmark tests, 1.6b rivals 4b general models, while 450m is equivalent to a 2b model.
Weights are now available for download on Hugging Face.