gliner2-base-v1

About GLiNER2#

GLiNER2 is a unified multi-task model that performs named entity recognition, zero-shot classification, relation extraction, and structured JSON extraction — all from a single model. It uses a DeBERTa-v3-base encoder with the GLiNER2 architecture.

Unlike traditional NER models that are limited to predefined entity types, GLiNER2 accepts arbitrary labels at inference time, making it extremely flexible for diverse extraction tasks.

GLiNER2 is a zero-shot model — you define the entity types, relation types, or classification labels at query time. No fine-tuning required.

Capabilities#

Named Entity Recognition: Extract entities with custom labels
Relation Extraction: Extract (subject, relation, object) triplets from text
Zero-Shot Classification: Classify text into arbitrary categories
JSON Extraction: Extract structured data matching a schema

Use Cases#

Use Case	Description
Document Processing	Extract people, organizations, dates from contracts or reports
Knowledge Graphs	Build knowledge graphs by extracting entities and their relationships
Content Tagging	Classify articles, support tickets, or reviews into custom categories
Data Extraction	Pull structured fields from unstructured text (e.g., invoices, resumes)

Antfly Inference API Usage#

Named Entity Recognition#

Extract entities with custom labels — no predefined entity types needed:

curl -X POST "http://localhost:8080/ai/v1/recognize" \
-H "Content-Type: application/json" \
-d '{
"model": "gliner2-base-v1",
"texts": ["Steve Jobs founded Apple in Cupertino, California."],
"labels": ["person", "organization", "location"]
}'

import requests

response = requests.post(
  "http://localhost:8080/ai/v1/recognize",
  json={
      "model": "gliner2-base-v1",
      "texts": ["Steve Jobs founded Apple in Cupertino, California."],
      "labels": ["person", "organization", "location"]
  }
)

entities = response.json()["entities"]
# [{"text": "Steve Jobs", "label": "person", "score": 0.98}, ...]

const response = await fetch("http://localhost:8080/ai/v1/recognize", {
method: "POST",
headers: { "Content-Type": "application/json" },
body: JSON.stringify({
  model: "gliner2-base-v1",
  texts: ["Steve Jobs founded Apple in Cupertino, California."],
  labels: ["person", "organization", "location"]
})
});

const { entities } = await response.json();

Relation Extraction#

Extract relationships between entities as (subject, relation, object) triplets:

curl -X POST "http://localhost:8080/ai/v1/recognize" \
-H "Content-Type: application/json" \
-d '{
"model": "gliner2-base-v1",
"texts": ["Steve Jobs founded Apple in 1976."],
"relation_labels": ["founded by", "located in", "born in"]
}'

import requests

response = requests.post(
  "http://localhost:8080/ai/v1/recognize",
  json={
      "model": "gliner2-base-v1",
      "texts": ["Steve Jobs founded Apple in 1976."],
      "relation_labels": ["founded by", "located in", "born in"]
  }
)

relations = response.json()["relations"]
# [[{"head": "Apple", "label": "founded by", "tail": "Steve Jobs", "score": 0.95}]]

const response = await fetch("http://localhost:8080/ai/v1/recognize", {
method: "POST",
headers: { "Content-Type": "application/json" },
body: JSON.stringify({
  model: "gliner2-base-v1",
  texts: ["Steve Jobs founded Apple in 1976."],
  relation_labels: ["founded by", "located in", "born in"]
})
});

const { relations } = await response.json();

Zero-Shot Classification#

Classify text into arbitrary categories:

curl -X POST "http://localhost:8080/ai/v1/classify" \
-H "Content-Type: application/json" \
-d '{
"model": "gliner2-base-v1",
"texts": ["Apple reported record quarterly revenue of $124 billion."],
"labels": ["finance", "technology", "sports", "politics"]
}'

import requests

response = requests.post(
  "http://localhost:8080/ai/v1/classify",
  json={
      "model": "gliner2-base-v1",
      "texts": ["Apple reported record quarterly revenue of $124 billion."],
      "labels": ["finance", "technology", "sports", "politics"]
  }
)

classifications = response.json()["classifications"]
# [{"label": "finance", "score": 0.92}, {"label": "technology", "score": 0.87}, ...]

const response = await fetch("http://localhost:8080/ai/v1/classify", {
method: "POST",
headers: { "Content-Type": "application/json" },
body: JSON.stringify({
  model: "gliner2-base-v1",
  texts: ["Apple reported record quarterly revenue of $124 billion."],
  labels: ["finance", "technology", "sports", "politics"]
})
});

const { classifications } = await response.json();

JSON Extraction#

Extract structured data from text matching a schema:

curl -X POST "http://localhost:8080/ai/v1/extract" \
-H "Content-Type: application/json" \
-d '{
"model": "gliner2-base-v1",
"texts": ["John Smith is a 35-year-old engineer at Google in Mountain View."],
"schema": {
  "person": ["name::str", "age::str", "job_title::str", "company::str", "city::str"]
}
}'

import requests

response = requests.post(
  "http://localhost:8080/ai/v1/extract",
  json={
      "model": "gliner2-base-v1",
      "texts": ["John Smith is a 35-year-old engineer at Google in Mountain View."],
      "schema": {
          "person": ["name::str", "age::str", "job_title::str", "company::str", "city::str"]
      }
  }
)

results = response.json()["results"]
# [[{"name": "John Smith", "age": "35", "job_title": "engineer", "company": "Google", "city": "Mountain View"}]]

const response = await fetch("http://localhost:8080/ai/v1/extract", {
method: "POST",
headers: { "Content-Type": "application/json" },
body: JSON.stringify({
  model: "gliner2-base-v1",
  texts: ["John Smith is a 35-year-old engineer at Google in Mountain View."],
  schema: {
    person: ["name::str", "age::str", "job_title::str", "company::str", "city::str"]
  }
})
});

const { results } = await response.json();

Supported Tasks#

Task	Endpoint	Description
NER (default)	`/recognize` with `labels`	Extract entities with custom labels
Relations	`/recognize` with `relation_labels`	Extract (subject, relation, object) triplets
Classification	`/classify` with `labels`	Zero-shot text classification
JSON Extraction	`/extract` with `schema`	Extract structured fields from text

GLiNER2 has a 512 token context limit. For longer documents, split into chunks before processing.

Antfly Configuration#

To use GLiNER2 as a recognizer in your Antfly configuration:

recognizer:
  provider: antfly
  model: gliner2-base-v1

For tables with entity extraction:

tables:
  documents:
    indexes:
      entities:
        type: ner
        recognizer:
          provider: antfly
          model: gliner2-base-v1
        labels:
          - person
          - organization
          - location
          - date

Tag	Status
latest	Available
v1.0.0	Latest

About GLiNER2#

Capabilities#

Use Cases#

Antfly Inference API Usage#

Named Entity Recognition#

Relation Extraction#

Zero-Shot Classification#

JSON Extraction#

Supported Tasks#

Antfly Configuration#

Versions

Usage

Pull the model

Start Antfly Inference

Use in configuration

Available variants (smaller/faster)