# mmBERT-32K PII Detector (LoRA)

Token classification model for detecting Personally Identifiable Information (PII) based on **mmBERT-32K-YaRN**.

## Model Description

This model performs Named Entity Recognition (NER) to identify and classify PII entities in text.

### Supported PII Types (17 categories)
- **PERSON**: Names of individuals
- **ORGANIZATION**: Company/organization names
- **STREET_ADDRESS**: Physical addresses
- **GPE**: Geopolitical entities (cities, countries)
- **EMAIL_ADDRESS**: Email addresses
- **PHONE_NUMBER**: Phone numbers
- **DATE_TIME**: Dates and times
- **CREDIT_CARD**: Credit card numbers
- **US_SSN**: US Social Security Numbers
- **IP_ADDRESS**: IP addresses
- **DOMAIN_NAME**: Domain names
- **IBAN_CODE**: International Bank Account Numbers
- **US_DRIVER_LICENSE**: US driver's license numbers
- **ZIP_CODE**: ZIP/postal codes
- **AGE**: Age information
- **NRP**: Nationality, religion, political group
- **TITLE**: Titles and honorifics

### Base Model
- **Base**: [llm-semantic-router/mmbert-32k-yarn](https://huggingface.co/llm-semantic-router/mmbert-32k-yarn)
- **Architecture**: ModernBERT with YaRN RoPE scaling
- **Context Length**: 32,768 tokens
- **Task**: Token Classification (BIO tagging)

### Training Details
- **Method**: LoRA fine-tuning
- **LoRA Rank**: 8
- **Dataset**: Presidio synthetic PII dataset
- **Labels**: 35 (17 entity types × 2 BIO tags + O)

## Usage

```python
from transformers import AutoTokenizer, AutoModelForTokenClassification
from peft import PeftModel

base_model = "llm-semantic-router/mmbert-32k-yarn"
adapter = "llm-semantic-router/mmbert32k-pii-detector-lora"

tokenizer = AutoTokenizer.from_pretrained(adapter)
model = AutoModelForTokenClassification.from_pretrained(base_model, num_labels=35)
model = PeftModel.from_pretrained(model, adapter)

text = "Contact John Smith at john@example.com or 555-1234"
inputs = tokenizer(text, return_tensors="pt")
outputs = model(**inputs)
```

## Intended Use

- PII detection and redaction
- Data privacy compliance
- Sensitive information filtering
- GDPR/CCPA compliance tools

## License

Apache 2.0