OpenMed-PII-Bengali-mSuperClinical-Large-279M-v1

What it does

This model detects 54 types of personal identifiers, including names, addresses, and financial details, in Bengali clinical and biomedical text. It is a 278M-parameter DeBERTa-v3 model with a reported F1 score of 0.857 on the AI4Privacy and synthetic Bengali PII test set.

Best for

• De-identifying Bengali medical records
• Removing personal data from clinical notes
• Anonymizing Bengali biomedical research documents

What it detects (54 labels)

• Account name
• Age
• Amount
• Bank account
• BIC code
• Bitcoin address
• Building number
• City
• County
• Credit card number
• Credit card issuer
• Currency
• Currency code
• Currency name
• Currency symbol
• Card CVV
• Date
• Date of birth
• Email
• Ethereum address
• Eye color
• First name
• Gender
• GPS coordinates
• Height
• IBAN
• IMEI
• IP address
• Job department
• Job title
• Last name
• Litecoin address
• MAC address
• Masked number
• Middle name
• Occupation
• Ordinal direction
• Organization
• Password
• Phone
• PIN
• Prefix
• Secondary address
• Sex
• Social security number
• State
• Street
• Time
• URL
• User agent
• Username
• Vehicle identification number
• Vehicle registration
• ZIP code

Quality

This model was trained on a combination of: - AI4Privacy PII Masking 200K: Multilingual base dataset (200K records across 8 languages) - NVIDIA Nemotron-PII: Seed dataset for synthetic data generation - Synthetic Bengali Data: ~25K high-quality samples generated with locale-specific formatting (NID format, +880 phones, Bengali names, ৳ currency)

Use it

pip install transformers torch

from transformers import pipeline
ner = pipeline("token-classification", model="OpenMed/OpenMed-PII-Bengali-mSuperClinical-Large-279M-v1", aggregation_strategy="simple")
print(ner("Patient prescribed 10 mg aspirin for hypertension."))

pip install --upgrade "openmed[onnx-runtime]"

from openmed import OnnxModel
model = OnnxModel.from_pretrained("OpenMed/OpenMed-PII-Bengali-mSuperClinical-Large-279M-v1-onnx-android")
for entity in model("The patient started imatinib after the consultation."):
    print(entity.to_dict())

pip install "openmed[mlx]"

from openmed import analyze_text
from openmed.core.config import OpenMedConfig
result = analyze_text("Patient started imatinib in March.", model_name="OpenMed/OpenMed-PII-Bengali-mSuperClinical-Large-279M-v1", config=OpenMedConfig(backend="mlx"))
for e in result.entities:
    print(e.label, e.text)