GLiNER2 base

GLiNER2 base

Extracts entities, classifies text, pulls out structured data and finds relations in plain text, with one small model that runs on a CPU. It suits developers who want information extraction without a GPU or an outside API.

Made by fastino on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: GLiNER2 base is an information extraction model from fastino. A single model handles four jobs. It finds named entities, labels whole texts, pulls fields into a structure and links entities through relations. It was made so you do not need a chain of separate tools or an outside API for them.

You describe what you want, such as a list of entity labels or a set of fields, and it reads your text for those. The card says it is built for efficient CPU inference.

Who it suits: Developers and data teams who need to turn free text into labels and fields suit it best. It runs from Python.

Industry | Who uses it | How they use it | Example
Software | Backend developers | Pull people, companies and products out of text | Extract entities from a news line
Customer support | Support teams | Sort comments by sentiment or topic | Label a review as positive or negative
E-commerce | Catalog teams | Turn product descriptions into fields | Read name, storage and price from a listing

Anyone who wants a chat assistant can skip it, because it reads text and returns structure, not conversation.

Where to use it: You run it yourself in Python, on Windows, macOS or Linux.

• Python scripts that import the gliner2 package
• Windows, macOS or Linux with Python installed
• Machines without a GPU, since the card says it is made for CPU use

Use case | Who | What they get back
Entity extraction | Developers | A dictionary of entities grouped under your labels
Text classification | Support teams | A single label or several labels for each text
Structured extraction | Catalog teams | Fields such as name, storage and price

How to set it up: You need Python and the gliner2 package. The model downloads when you load it by name.

• Python on Windows, macOS or Linux
• The gliner2 package, installed with pip
• Disk space for about 834 MB of weights in F32

1. Install Python.
2. Install the gliner2 package with pip.
3. Optionally download the model files ahead of time with the hub tools.
4. Load the model by its name in a Python script.

pip install gliner2

pip install -U huggingface_hub
hf download fastino/gliner2-base-v1

How to use it: Load the model, then call the method for the job you want. For entities you pass the text and a list of labels.

from gliner2 import GLiNER2

# Load the model
extractor = GLiNER2.from_pretrained("fastino/gliner2-base-v1")

# Extract entities
text = "Apple CEO Tim Cook announced iPhone 15 in Cupertino yesterday."
result = extractor.extract_entities(text, ["company", "person", "product", "location"])

print(result)
# Output: {'entities': {'company': ['Apple'], 'person': ['Tim Cook'], 'product': ['iPhone 15'], 'location': ['Cupertino']}}

For classification you pass the text and the labels you allow.

# Single-label classification
result = extractor.classify_text(
    "This laptop has amazing performance but terrible battery life!",
    {"sentiment": ["positive", "negative", "neutral"]}
)
print(result)
# Output: {'sentiment': 'negative'}

Tool | What it does
extract_entities | Finds entities for the labels you give
classify_text | Picks one label, or several, for a text
extract_json | Fills the fields you describe from a text
create_schema | Builds one schema that combines several tasks
extract | Runs a combined schema over a text

Example: You pass a sentence about Apple's CEO Tim Cook announcing a phone in Cupertino, with the labels company, person, product and location. It returns Apple as the company, Tim Cook as the person, the phone as the product and Cupertino as the location.

For newcomers and experts: Think of the model as a reader that marks up a text for you. You say what to look for, and it hands back labels and fields. Your first small try is the entity example above with a sentence of your own.

Under the hood, one model handles four tasks in a single forward pass, and you steer it with labels or a schema at run time. The card describes about 205M parameters and CPU-first inference with local processing. The trade-off is that the labels you write decide what you get back.

Your text and your labels or schema
One model reads them together
Entities, classes, fields or relations
A plain Python dictionary comes back

Good to know: The card says processing stays local and needs no outside service once the model files are on your machine.

• Size: 208M parameters in the record, about 834 MB in F32, stored as Safetensors
• Language: English
• Other versions: Hugging Face lists 6 quantized and 12 fine-tuned versions made by other people
• License: apache-2.0
• Cost to run: none named
• Data it reaches: the text you give it, and the model files downloaded from Hugging Face

On Hugging Face, 13 November 2026

• Library: gliner2
• Parameters: 208M
• Weights: about 834 MB in F32; about 104 MB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 353,994
• Likes: 106
• License: apache-2.0
• Languages: en
• Quantized versions: 6
• Fine-tunes: 12
• Tags: gliner2, safetensors, extractor

Links

• Model: https://huggingface.co/fastino/gliner2-base-v1
• Files: https://huggingface.co/fastino/gliner2-base-v1/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download fastino/gliner2-base-v1

Then follow the usage in its model card. Nothing in this kit runs by itself.