traromal/AIccel_entity_masker_Gemma3_270m

💡 Example Inference Code

You can try this PII Masking model directly with the following script:

from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer
import torch

# ----------------------------
# Load model & tokenizer
# ----------------------------
model_name = "traromal/AIccel_entity_masker_Gemma3_270m"
print(f"Loading model: {model_name}")

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    trust_remote_code=True,
    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
    low_cpu_mem_usage=True
)

device = "cuda" if torch.cuda.is_available() else "cpu"
model = model.to(device)
print(f"✅ Model loaded on {device}")

# ----------------------------
# System prompt
# ----------------------------
SYSTEM_PROMPT = """You are a global data privacy expert.
Identify and mask all PII (Personally Identifiable Information) in text.
Replace each with an appropriate tag like [NAME], [AADHAR_NUMBER], [PHONE], etc.
Also list detected entities with their type and sensitivity level."""

# ----------------------------
# Masking function
# ----------------------------
def mask_pii(text, stream=False):
    messages = [
        {'role': 'system', 'content': SYSTEM_PROMPT},
        {'role': 'user', 'content': f'Mask all sensitive PII in:\n\n"{text}"'}
    ]

    chat = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
    inputs = tokenizer(chat, return_tensors="pt").to(device)

    if stream:
        streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
        model.generate(
            **inputs,
            max_new_tokens=512,
            temperature=0.7,
            top_p=0.9,
            top_k=50,
            do_sample=True,
            streamer=streamer,
            pad_token_id=tokenizer.pad_token_id,
            eos_token_id=tokenizer.eos_token_id,
        )
    else:
        outputs = model.generate(
            **inputs,
            max_new_tokens=512,
            temperature=0.7,
            top_p=0.9,
            top_k=50,
            do_sample=True,
            pad_token_id=tokenizer.pad_token_id,
            eos_token_id=tokenizer.eos_token_id,
        )
        response = tokenizer.decode(outputs[0], skip_special_tokens=False)
        if "<start_of_turn>model" in response:
            response = response.split("<start_of_turn>model")[-1].replace("<end_of_turn>", "").strip()
        print(response)
        return response

# ----------------------------
# Quick examples
# ----------------------------
examples = [
    "My name is Rajesh Kumar and my Aadhar number is 1234-5678-9012. Contact me at +91-9876543210.",
    "Patient Priya Sharma, Blood Group: B+, UHID: MH2023-12345, DOB: 15/08/1990.",
    "Please transfer ₹50,000 to account 123456789012 (IFSC: HDFC0001234). UPI ID: amit.kumar@paytm.",
    "John Smith, SSN: 123-45-6789, email: john.smith@gmail.com",
]

for text in examples:
    print("\n🧩 Original:", text)
    print("🔒 Masked:")
    mask_pii(text, stream=True)
    print("=" * 80)

# ----------------------------
# Interactive mode
# ----------------------------
while True:
    user_text = input("\n🔒 Enter text to mask (or 'exit'): ").strip()
    if user_text.lower() in ["exit", "quit", "q"]:
        break
    mask_pii(user_text, stream=True)