The compact model GLiNER2.5 Small for data extraction has been released

Developers presented GLiNER2.5 Small — a specialized model for unified information extraction based on schemes. The release is focused on tasks that require high processing speed and the possibility of local deployment. The model's architecture is based on 74 million parameters, which allows it to be used effectively on consumer equipment. The main specialization of the system is working with English-language texts, including entity extraction, classification, and determining relationships between objects. The tool supports work through the AutoExtractor system, which automates the process of structuring unmarked data. The model is optimized for local inference, which eliminates the need to transfer confidential information to third-party servers. The compact size of the model provides low requirements for RAM and GPU computing power. This makes the solution suitable for integration into edge devices and local data processing pipelines, where response latency is critical. The release of the Small version complements the GLiNER ecosystem, providing developers with a choice between the performance of large models and lightness for real-time tasks. The source code and model weights are already available for download in the repository on the Hugging Face platform.