Mindket

GLiNER2-PII

Newversion 1 · updated

Overview

Finds personal data such as names, emails, phone numbers and card numbers in text across seven languages, using 42 label types. It suits developers who build redaction and data-cleaning pipelines.

GLiNER2-PII is a kit in the Developer category, listed by Mindket. It is free. This is version 1, last updated on . Download it on this page, a zip of its files. Nothing in it runs by itself.

At a glance

You need
  • Python on Windows, macOS or Linux
  • The gliner2 package, installed with pip
  • Disk space for about 1.2 GB of weights in F32
Runs in
  • Python scripts that import the gliner2 package
  • Redaction and masking pipelines
  • Text in English, French, Spanish, German, Italian, Portuguese or Dutch
Start with
pip install gliner2

Platform

WindowsOperating system
macOSOperating system
LinuxOperating system

About

What it is

GLiNER2-PII takes fastino's GLiNER2 model and tunes it further for personal data. It detects personally identifiable information in text so you can mask it. It knows 42 entity types across 7 languages, and the model was trained on a synthetic corpus of 4,910 annotated texts.

You choose which labels to look for each time you run it. The model conditions on the labels you give, so any subset of the 42 works.

Who it suits

Developers and data teams who must find or hide personal data in text suit it best. It runs from Python.

IndustryWho uses itHow they use itExample
ComplianceData governance teamsFind personal data for GDPR or CCPA policy workFlag names, emails and IDs in documents
Machine learningData engineersClean text before it becomes training dataMask the exact spans of personal data
Legal and medicalRedaction teamsHide sensitive spans in documentsMask phone numbers and names in a letter

Anyone who needs a guarantee of compliance should not rely on it alone, because its own benchmark average F1 is 0.477.

Where to use it

You run it yourself in Python, on Windows, macOS or Linux.

  • Python scripts that import the gliner2 package
  • Redaction and masking pipelines
  • Text in English, French, Spanish, German, Italian, Portuguese or Dutch
Use caseWhoWhat they get back
PII redactionRedaction teamsExact character spans to mask
Policy routingGovernance teamsFine-grained types that map to your own rules
Training-data hygieneData engineersText with personal data masked before training
Multi-language pipelinesGlobal teamsOne model for seven languages

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 1.2 GB 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-privacy-filter-PII-multi

How to use it

Load the model, give it a text and the labels you want, and call extract_entities. The options below ask for a confidence score and the text span of each match.

from gliner2 import GLiNER2

model = GLiNER2.from_pretrained("fastino/gliner2-privacy-filter-PII-multi")

text = "Email john.smith@acme.com or call +1 415 555 0199."
labels = ["email", "phone_number", "person"]

result = model.extract_entities(
    text,
    labels,
    threshold=0.5,
    include_confidence=True,
    include_spans=True,
)

print(result)
A few of the 42 labels
GroupExample labels
Namesperson, full_name, date_of_birth
Contactemail, phone_number, postal_code
Government IDspassport_number, tax_id
Bankingiban, card_number
Secretspassword, api_key, access_token

Example

You pass a short message that holds an email address and a phone number, with the labels email, phone_number and person. It returns the matches it finds for those labels, each with a confidence score and its text span.

For newcomers and experts

Think of the model as a highlighter for private details in text. You say which kinds of detail to mark, and it marks them. Your first small try is the quick start above with a sentence of your own.

Under the hood, it is GLiNER2 fine-tuned on synthetic data, and the labels you pass at run time steer what it extracts. The card reports the highest span-level F1 among four compared systems on the SPY benchmark, with recall of 0.718 on average among GLiNER-based detectors.

  1. Your text and a subset of the 42 labels
  2. The model conditions on those labels
  3. Matches come back with spans and scores
  4. You mask the spans you want hidden

Good to know

The benchmark numbers come from the model card and cover legal and medical text only.

  • Size: 307M parameters in the record, about 1.2 GB in F32, stored as Safetensors
  • Languages: English, French, Spanish, German, Italian, Portuguese and Dutch
  • Other versions: Hugging Face lists 11 quantized and 8 fine-tuned versions made by other people
  • License: apache-2.0
  • Cost to run: none named
  • Data it reaches: the text you give it, which may hold personal data, and the model files downloaded from Hugging Face
  • Benchmark: the card's table gives an average F1 of .477 for this model, against .400, .398 and .380 for three other systems

README and files

README

GLiNER2-PII

Finds personal data such as names, emails, phone numbers and card numbers in text across seven languages, using 42 label types. It suits developers who build redaction and data-cleaning pipelines.

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-PII takes fastino's GLiNER2 model and tunes it further for personal data. It detects personally identifiable information in text so you can mask it. It knows 42 entity types across 7 languages, and the model was trained on a synthetic corpus of 4,910 annotated texts.

You choose which labels to look for each time you run it. The model conditions on the labels you give, so any subset of the 42 works.

Who it suits: Developers and data teams who must find or hide personal data in text suit it best. It runs from Python.

Industry | Who uses it | How they use it | Example
Compliance | Data governance teams | Find personal data for GDPR or CCPA policy work | Flag names, emails and IDs in documents
Machine learning | Data engineers | Clean text before it becomes training data | Mask the exact spans of personal data
Legal and medical | Redaction teams | Hide sensitive spans in documents | Mask phone numbers and names in a letter

Anyone who needs a guarantee of compliance should not rely on it alone, because its own benchmark average F1 is 0.477.

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

  • Python scripts that import the gliner2 package
  • Redaction and masking pipelines
  • Text in English, French, Spanish, German, Italian, Portuguese or Dutch
Use case | Who | What they get back
PII redaction | Redaction teams | Exact character spans to mask
Policy routing | Governance teams | Fine-grained types that map to your own rules
Training-data hygiene | Data engineers | Text with personal data masked before training
Multi-language pipelines | Global teams | One model for seven languages

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 1.2 GB 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-privacy-filter-PII-multi

How to use it: Load the model, give it a text and the labels you want, and call extract_entities. The options below ask for a confidence score and the text span of each match.

from gliner2 import GLiNER2

model = GLiNER2.from_pretrained("fastino/gliner2-privacy-filter-PII-multi")

text = "Email john.smith@acme.com or call +1 415 555 0199."
labels = ["email", "phone_number", "person"]

result = model.extract_entities(
    text,
    labels,
    threshold=0.5,
    include_confidence=True,
    include_spans=True,
)

print(result)
Title: A few of the 42 labels
Group | Example labels
Names | person, full_name, date_of_birth
Contact | email, phone_number, postal_code
Government IDs | passport_number, tax_id
Banking | iban, card_number
Secrets | password, api_key, access_token

Example: You pass a short message that holds an email address and a phone number, with the labels email, phone_number and person. It returns the matches it finds for those labels, each with a confidence score and its text span.

For newcomers and experts: Think of the model as a highlighter for private details in text. You say which kinds of detail to mark, and it marks them. Your first small try is the quick start above with a sentence of your own.

Under the hood, it is GLiNER2 fine-tuned on synthetic data, and the labels you pass at run time steer what it extracts. The card reports the highest span-level F1 among four compared systems on the SPY benchmark, with recall of 0.718 on average among GLiNER-based detectors.

Your text and a subset of the 42 labels
The model conditions on those labels
Matches come back with spans and scores
You mask the spans you want hidden

Good to know: The benchmark numbers come from the model card and cover legal and medical text only.

  • Size: 307M parameters in the record, about 1.2 GB in F32, stored as Safetensors
  • Languages: English, French, Spanish, German, Italian, Portuguese and Dutch
  • Other versions: Hugging Face lists 11 quantized and 8 fine-tuned versions made by other people
  • License: apache-2.0
  • Cost to run: none named
  • Data it reaches: the text you give it, which may hold personal data, and the model files downloaded from Hugging Face
  • Benchmark: the card's table gives an average F1 of .477 for this model, against .400, .398 and .380 for three other systems

On Hugging Face, 16 November 2026

  • Task: token classification
  • Library: gliner2
  • Parameters: 307M
  • Weights: about 1.2 GB in F32; about 154 MB at 4-bit (estimate)
  • Formats: Safetensors
  • Downloads in the last 30 days: 132,410
  • Likes: 115
  • License: apache-2.0
  • Languages: en, fr, es, de, it, pt, nl
  • Quantized versions: 11
  • Fine-tunes: 8
  • Tags: gliner2, safetensors, extractor, privacy, redaction, information-extraction, span-extraction, token-classification

Links

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download fastino/gliner2-privacy-filter-PII-multi

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

Files
README.md  6 KB

1 file, 6 KB, downloaded as a zip.

What’s new

  • Version 1 · 2026-11-15

Compare with similar

Listed Hugging Face models in the same category that share the most tags with this one.

ProductDownloads, last 30 daysLikesParametersLicenseLast change
GLiNER2-PII This page132,410115307Mapache-2.0
UltraBERT-150M-clairic-extractionShares span-extraction · information-extraction · token-classification1385151Mapache-2.0
cbx-indic-pii-extractor-58MShares redaction · privacy · token-classification96457.6Mapache-2.0
cbx-indic-pii-extractor-17MShares redaction · privacy · token-classification76316.7Mapache-2.0
GLiNER2 baseShares extractor · gliner2 · safetensors353,994106208Mapache-2.0
GLiNER2.5 BaseShares information-extraction · gliner2 · safetensors132,02743194Mapache-2.0
Numbers as Mindket last read them, from 2026-11-13 to 2026-11-15.

Reviews

No ratings yet

Write a review

No reviews yet.

More from Mindket

See all from Mindket

all-mpnet-base-v2

sentence-transformers · Kit

Turns sentences and paragraphs into 768-number vectors so you can search, cluster and compare text by meaning. It suits developers and data teams working in En…

New

multilingual-e5-base

intfloat · Kit

Turns text in about 100 languages into vectors so you can compare meaning across them, for search and matching. Made for developers who build multilingual retr…

New

sprite-gen

aldegad · Skill

Turns one base drawing of a character into game-ready sprites, either an atlas with a manifest or transparent motion loops, for game developers working in Code…

New4097

Similar in Developer

See all in Developer

VibeSkills

foryourhealth111-pixel · Skill

Picks the right skills from a large installed pool and breaks each task into checkable steps for your coding agent. In its own benchmark, mean reward rose 21.1…

New3821

claude-bughunter

elementalsouls · Skill

Gives Claude Code 83 skills and 15 slash commands for bug bounty hunting and authorized red-team work, from web vulnerability classes to enterprise identity at…

New6853

Runs Qwen3.8-27B with a vocabulary cut down to ASCII, so a single 16 GB graphics card can hold a much longer prompt. A GGUF build for English text and code wit…

New