README.md

metadata

title: Multi-Model Indian Address NER
emoji: 🏠
colorFrom: blue
colorTo: green
sdk: gradio
sdk_version: 5.35.0
app_file: app.py
pinned: false

Multi-Model Indian Address NER Demo

This is a Gradio-based demo that allows you to compare three different Indian Address NER models:

• TinyBERT - Lightweight and fast
• ModernBERT - Modern architecture
• IndicBERT - Indic language optimized

What it does

This application allows you to:

1. Single Model Analysis: Choose one model and extract entities from Indian addresses
2. Model Comparison: Compare how all three models perform on the same address
3. Interactive Testing: Use sample addresses or input your own

The models can identify:

• Building names
• Floor numbers
• House details
• Roads
• Sub-localities
• Localities
• Landmarks
• Cities
• States
• Countries
• Pincodes

How to use

Single Model Analysis

1. Select a model from the dropdown (TinyBERT, ModernBERT, or IndicBERT)
2. Enter an Indian address in the text box
3. Click "Extract Entities" or press Enter
4. View the extracted entities with confidence scores

Model Comparison

1. Go to the "Model Comparison" tab
2. Enter an address
3. Click "Compare All Models"
4. See how each model performs on the same input

Example addresses

• Shop No 123, Sunshine Apartments, Andheri West, Mumbai, 400058
• DLF Cyber City, Sector 25, Gurgaon, Haryana
• Flat 201, MG Road, Bangalore, Karnataka, 560001

Model Information

TinyBERT

• Parameters: ~66.4M
• Advantages: Fastest inference, lowest memory
• Best for: Real-time applications, mobile deployment

ModernBERT

• Parameters: ~599MB model
• Advantages: Modern architecture, balanced performance
• Best for: High accuracy with reasonable speed

IndicBERT

• Parameters: ~131MB model
• Advantages: Optimized for Indian languages/contexts
• Best for: Mixed language addresses, regional contexts

Framework: PyTorch + Transformers