gliner2-pii500k-final-model

Fine-tuned GLiNER2 for PII (Personally Identifiable Information) extraction.
Model type: extractor

• 44 entity types across 8 PII taxonomy groups
• Languages: English · German · Vietnamese
• Uploaded: 2026-07-20

Entity Labels

Group	Labels
Identity	PREFIX · PERSON · GENDER · AGE · BIRTHDATE
Contact & Location	PHONE · EMAIL · LOCATION · ADDRESS · ZIP_CODE · COORDINATE
Digital & Network	USERNAME · ACCOUNT_ID · TICKET_ID · PASSWORD · PIN · API_KEY · URL · IP
Financial	BANK_ACCOUNT · MONEY · CARD_ISSUER · CARD_NUMBER · CVV · IBAN · SWIFT · WALLET
Employment	JOB_TITLE · ORGANIZATION · EMPLOYEE_ID
Government & Legal	NATIONAL_ID · PASSPORT · LICENSE · PLATE · TIN
Time	DATE · TIME
Sensitive	MARITAL · RELIGION · ETHNICITY · TRADE_UNION · NATIONALITY · INSURANCE_ID · MEDICAL_INFO

Training Configuration

Parameter	Value
Epochs	6
Batch size	16
Encoder LR	1e-05
Task LR	0.0005
Max seq len	2048
Precision	fp16

Usage

from gliner2 import GLiNER2

model = GLiNER2.from_pretrained("AITeamUIT/gliner2-pii500k-final-model")

text = "Please contact John Doe at john.doe@example.com or call +1-555-234-5678."
labels = ['PERSON', 'DATE', 'JOB_TITLE', 'TIME', 'LOCATION', 'PREFIX', 'ORGANIZATION', 'EMAIL', 'PHONE', 'ADDRESS', 'GENDER', 'COORDINATE', 'MONEY', 'IP', 'ZIP_CODE', 'NATIONALITY', 'AGE', 'USERNAME', 'URL', 'MARITAL', 'TRADE_UNION', 'BIRTHDATE', 'CARD_ISSUER', 'PIN', 'PASSPORT', 'MEDICAL_INFO', 'IBAN', 'PASSWORD', 'ETHNICITY', 'RELIGION', 'EMPLOYEE_ID', 'INSURANCE_ID', 'TIN', 'SWIFT', 'CARD_NUMBER', 'ACCOUNT_ID', 'WALLET', 'NATIONAL_ID', 'CVV', 'BANK_ACCOUNT', 'LICENSE', 'TICKET_ID', 'PLATE', 'API_KEY']
label_descriptions = {'PERSON': 'Person names', 'DATE': 'Calendar dates', 'JOB_TITLE': 'Job titles', 'TIME': 'Time values', 'LOCATION': 'General locations', 'PREFIX': 'Titles before names', 'ORGANIZATION': 'Organization names', 'EMAIL': 'Email addresses', 'PHONE': 'Phone numbers', 'ADDRESS': 'Street addresses', 'GENDER': 'Sex or gender', 'COORDINATE': 'Geographic coordinates', 'MONEY': 'Monetary amounts', 'IP': 'IP addresses', 'ZIP_CODE': 'Postal or ZIP codes', 'NATIONALITY': 'Legal nationality', 'AGE': 'Age values', 'USERNAME': 'Usernames or handles', 'URL': 'Web links or URLs', 'MARITAL': 'Marital status', 'TRADE_UNION': 'Trade union membership', 'BIRTHDATE': 'Date of birth', 'CARD_ISSUER': 'Card brands or issuers', 'PIN': 'Authentication PINs', 'PASSPORT': 'Passport numbers', 'MEDICAL_INFO': 'Health or medical information', 'IBAN': 'International bank account numbers', 'PASSWORD': 'Account passwords', 'ETHNICITY': 'Ethnic or racial group', 'RELIGION': 'Religious affiliation', 'EMPLOYEE_ID': 'Employee identifiers', 'INSURANCE_ID': 'Insurance identifiers', 'TIN': 'Tax identification numbers', 'SWIFT': 'Bank SWIFT or BIC codes', 'CARD_NUMBER': 'Payment card numbers', 'ACCOUNT_ID': 'Account identifiers', 'WALLET': 'Crypto wallet addresses', 'NATIONAL_ID': 'National ID numbers', 'CVV': 'Card security codes', 'BANK_ACCOUNT': 'Bank account numbers', 'LICENSE': 'Driver license numbers', 'TICKET_ID': 'Support or case IDs', 'PLATE': 'Vehicle plate numbers', 'API_KEY': 'API keys or tokens'}

entities = model.extract(text, labels=labels, label_descriptions=label_descriptions, threshold=0.5)
for ent in entities:
    print(ent)

Limitations

• Span boundaries may be imperfect for entities crossing chunking boundaries.
• High-precision structured fields (EMAIL, PHONE, IBAN, etc.) benefit from the bundled regex rule-detector as a gap-filler.