---
license: mit
tags:
- seq2seq
- bart
- lora
- pii-masking
- peft
- transformer
model-index:
- name: Mercy-62/bart-base-pii-masker-lora
  results:
  - task:
      name: Text Anonymization (PII Masking)
      type: text-anonymization
    metrics:
    - type: validation_loss
      value: 0.38088
datasets:
- ai4privacy/pii-masking-200k
---

# 🧠 Mercy-62/bart-base-pii-masker-lora

A **LoRA fine-tuned BART model** for **PII (Personally Identifiable Information) masking** transforms sensitive text into privacy-safe form while keeping sentence meaning intact.

---

## ✨ Overview

`Mercy-62/bart-base-pii-masker-lora` is built on top of [`facebook/bart-base`](https://huggingface.co/facebook/bart-base) and fine-tuned using the [PEFT](https://github.com/huggingface/peft) LoRA approach for **lightweight adaptation**.  

It is designed to automatically detect and mask **personally identifiable information** such as:
- Names  
- Emails  
- Phone numbers  
- CNICs / National IDs  
- Bank account numbers  
- Dates  
- Salary or monetary values  

---

## 💡 Example

**Input:**
```
"Customer John Smith applied for a credit card on May 10, 2023."
```
**Output:**
```
Customer [NAME] applied for a credit card on [DATE].
```

---

## 🧩 Model Details

| Component | Value |
|------------|--------|
| Base Model | `facebook/bart-base` |
| Adapter Type | LoRA (Low-Rank Adaptation) |
| Frameworks | PyTorch, Transformers, PEFT |
| Task | Text-to-Text Generation |
| Language | English |

---

## 🧪 Example Inference

```python
from peft import PeftModel
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer

# Load model
base_model = AutoModelForSeq2SeqLM.from_pretrained("facebook/bart-base")
model = PeftModel.from_pretrained(base_model, "Mercy-62/bart-base-pii-masker-lora")
tokenizer = AutoTokenizer.from_pretrained("facebook/bart-base")

def test_pii_masking(model, tokenizer, samples):
    device = model.device
    for text in samples:
        inputs = tokenizer(text, return_tensors="pt", truncation=True).to(device)
        outputs = model.generate(**inputs, max_length=80)
        print(f"\nInput: {text}")
        print("Output:", tokenizer.decode(outputs[0], skip_special_tokens=True))

pii_samples = [
    "Customer John Smith applied for a credit card on May 10, 2023.",
    "Contact number: +92-300-1234567, Email: sara.khan@gmail.com",
]

test_pii_masking(model, tokenizer, pii_samples)