PII Guard DistilBERT

Fine-tuned DistilBERT-multilingual for Russian PII detection. 21 entity types in BIO format. F1=0.9938 on pii_benchmark.

Usage

from transformers import AutoModelForTokenClassification, AutoTokenizer
model = AutoModelForTokenClassification.from_pretrained("astralinux/pii-guard-distilbert")
tokenizer = AutoTokenizer.from_pretrained("astralinux/pii-guard-distilbert")

Entity Types

Tag Description
B-FIRST_NAME First name
B-LAST_NAME Last name
B-MIDDLE_NAME Middle name / patronymic
B-PHONE Phone number
B-EMAIL Email address
B-CITY City
B-STREET Street
B-HOUSE House number
B-COUNTRY Country
B-REGION Region
B-DISTRICT District
B-PASSPORT Passport number
B-SNILS SNILS
B-INN INN (tax ID)
B-OMS Medical insurance
B-CREDIT_CARD Credit card / bank account
B-DRIVER_LICENSE Driver license
B-MILITARY_ID Military ID
B-BIRTH_CERTIFICATE Birth certificate
B-URL URL
B-IP_ADDRESS IP address

Part of PII Guard

github.com/astralinux/pii-guard

Designed for 152-ФЗ compliance. Self-hosted. No data leaves your perimeter.

Downloads last month
44
Safetensors
Model size
0.1B params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support