GLiNER-2-XL

GLiNER 2 Overview

GLiNER 2 is Fastino’s open-source, schema-based information extraction model — a unified architecture for Named Entity Recognition (NER), Text Classification, and Structured Data Extraction in one forward pass.

It powers Fastino’s /gliner-2 inference API and can also be fine-tuned locally for custom domains such as finance, healthcare, law, and enterprise intelligence.

What is GLiNER 2?

GLiNER 2 (Generalized Language Interface for Named Entity Recognition v2) is a 205M–340M parameter model that unifies three traditionally separate tasks:

  1. Entity Extraction (NER)
    Identify key entities like people, organizations, products, and locations.

  2. Text Classification
    Assign labels (e.g., sentiment, category, intent) with single or multi-label support.

  3. Structured Data Extraction
    Parse complex fields into typed, hierarchical outputs (JSON-ready data).

Unlike legacy pipelines, GLiNER 2 performs all three tasks simultaneously through a single schema-driven prompt — improving efficiency, reducing latency, and ensuring consistency.

Why It Matters

Most extraction systems rely on large, fragmented LLM prompts or task-specific fine-tunes.
GLiNER 2 takes a different approach:

  • Unified Modeling: One encoder-decoder pipeline for all extraction tasks.

  • CPU-Optimized: Designed to run inference in real time without a GPU.

  • Schema-Driven: Extraction logic is declarative, not hard-coded.

  • Privacy-Safe: 100% local processing — no external dependencies.

  • LLM-Compatible: Outputs structured JSON easily consumed by downstream models or Fastino’s personalization APIs.

Key Features

Capability

Description

Multi-Task Learning

NER, classification, and structured extraction in one model.

Fast Inference

100ms–250ms latency on CPU for standard inputs.

Schema-Driven Prompts

Simple schema syntax defines output structure.

Local Privacy

No external calls or third-party dependencies.

Unified Output Format

Returns JSON outputs directly usable for LLM reasoning.

Open-Source + Hosted

Available on Hugging Face and Fastino’s inference endpoint.

Example Use Cases

  • Financial Reports: Extract trades, commissions, securities, and entities.

  • Healthcare: Identify patient details, symptoms, and prescriptions.

  • Legal Documents: Extract contracts, clauses, parties, and fees.

  • Product Reviews: Classify sentiment, extract product specs, and generate summaries.

  • Research Papers: Extract authors, affiliations, methods, and results.

Summary
GLiNER 2 is the foundation for unified schema-based information extraction — enabling real-time, structured understanding of text.
Available as both a hosted API (/gliner-2) and an open-source toolkit, it brings Fastino’s world-class personalization and extraction intelligence to any application.

On this page