mozgiDevelopers released the multilingual model GLiNER2.5 Multi
The development team introduced GLiNER2.5 Multi — an updated version of the model for unified information extraction. The system is designed to solve entity recognition tasks, text classification, and determining relationships between objects in a multilingual environment.
The model's architecture is based on the pre-trained encoder mDeBERTa-v3-base. The total number of parameters is 287 million, which provides a balance between performance and computational resource requirements. The model supports integration through the AutoExtractor library, simplifying deployment in existing data processing pipelines.
During internal testing, the system demonstrated high efficiency, scoring 96 points on the internal quality metric in just 3 hours of training. The developers focused on the tool's versatility, allowing users to extract structured data from unstructured texts without the need for fine-tuning for specific domains.
The release is available for download on the Hugging Face platform under an open license. Using GLiNER2.5 Multi enables the automation of entity extraction in projects that require support for multiple languages simultaneously, reducing the time spent setting up specific classifiers for each language.
