---
title: GLiNER2.5-base Information Extraction
emoji: 🔍
colorFrom: indigo
colorTo: purple
sdk: gradio
sdk_version: 6.25.0
app_file: app.py
short_description: Schema-driven NER, classification & relation extraction
python_version: "3.12"
startup_duration_timeout: 30m
---

# GLiNER2.5-base — Schema-Based Information Extraction

A demo of [`fastino/gliner2.5-base-v1`](https://huggingface.co/fastino/gliner2.5-base-v1),
a 194M-parameter unified model for:

- **Named Entity Recognition** — define your own entity types
- **Text Classification** — single- and multi-label, with task descriptions
- **Relation Extraction** — typed relations between entities

All tasks are zero-shot: you supply the schema (labels, descriptions, task
definitions) and the model does the rest — no fine-tuning needed.

## How it works

The model uses a **Boundary Extractor** architecture built on
`microsoft/deberta-v3-base`. It supports up to 4096 tokens and runs on GPU
via ZeroGPU.

## Usage

1. Pick a tab (Entity Extraction / Classification / Relations).
2. Enter your text.
3. Define your schema — entity types, classification tasks, or relation types.
4. Click the button. Entity extraction renders highlighted spans labelled
   with their entity type (and confidence); classification and relation
   extraction return JSON.