GLiNER2 Multi

GLiNER2 Multi is a compact information-extraction model that brings four common NLP tasks under one roof. According to its Hugging Face model card, it unifies Named Entity Recognition, Text Classification, Structured Data Extraction, and Relation Extraction into a single 205M-parameter architecture, letting practitioners run schema-based extraction and classification in one forward pass rather than chaining separate pipelines. The card also documents a straightforward Python interface through the gliner2 package, with usage examples such as extracting "company," "person," "product," and "location" entities from a sentence and performing single-label sentiment classification from a user-defined label set.

The model is positioned for lightweight, privacy-friendly deployment rather than large-scale serving on accelerators. Its creators emphasize CPU-first inference, describing it as lightning-fast on standard hardware with no GPU required, and highlight 100% local processing with zero external API dependencies, which makes it a practical fit for on-prem, edge, or sensitive-data workflows where sending text to a hosted API is undesirable. As a schema-driven zero-shot style extractor, it adapts to new label sets without retraining, suiting use cases like document tagging, lightweight relation mining, and structured data parsing where a small footprint, simple installation, and unified task handling matter more than raw scale.