mukuls9971/pii-deberta-v3-xsmall

PII Detection Model (DeBERTa-v3-xsmall, Fine-tuned)

This model is a fine-tuned version of microsoft/deberta-v3-xsmall for Named Entity Recognition (NER) focused on detecting Personally Identifiable Information (PII).

Model Details

• Base Model: microsoft/deberta-v3-xsmall
• Dataset: ai4privacy/pii-masking-200k (English subset, ~43k samples)
• Training: 5 epochs, batch size 32 (effective), learning rate 3e-5
• Best F1: 0.59

Intended Use

This model is designed for:

• Detecting PII entities in text (Names, Emails, Phone Numbers, Addresses, SSNs, etc.)
• Privacy compliance and data anonymization pipelines
• Research and development in privacy-preserving NLP

Limitations

• Trained primarily on English text.
• Performance may vary on domain-specific or out-of-distribution data.
• Not a replacement for comprehensive privacy review.

How to Use

from transformers import pipeline

pii_detector = pipeline("ner", model="mukuls9971/pii-deberta-v3-xsmall", aggregation_strategy="simple")
text = "Contact me at jane.smith@example.org or call (555) 123-4567."
results = pii_detector(text)
print(results)