---
title: Llama Address Intelligence
emoji: 🦙
colorFrom: purple
colorTo: pink
sdk: gradio
sdk_version: 4.44.0
app_file: app.py
pinned: false
license: apache-2.0
---

# Llama 3.2-1B Address Intelligence Demo

This Space demonstrates the capabilities of [shiprocket-ai/open-llama-1b-address-completion](https://huggingface.co/shiprocket-ai/open-llama-1b-address-completion), a fine-tuned Llama 3.2-1B model specialized for Indian address processing.

## What it does

This application showcases three main capabilities, with varying performance levels:

1. **Component Extraction** ⭐ **BEST PERFORMANCE** - Parse addresses into structured components (building, locality, pincode, etc.)
2. **Address Completion** ⚠️ **LIMITED** - Complete partial addresses (trained on limited data)
3. **Format Standardization** ⚠️ **LIMITED** - Convert informal addresses to standardized format (trained on limited data)

**Note**: This model performs best at **entity extraction**. The completion and standardization features have limited training data and may not always produce optimal results.

## Features

- **Lightweight**: Only 1.24B parameters for fast inference
- **Specialized**: Fine-tuned specifically for Indian address patterns
- **Versatile**: Handles multiple address intelligence tasks
- **Interactive**: Three separate tabs for different use cases
- **Real-time**: Optimized for quick responses

## How to use

### Component Extraction
1. Go to the "Extract Components" tab
2. Enter a complete address
3. Click "Extract Components" to see structured breakdown

### Address Completion
1. Go to the "Complete Address" tab  
2. Enter a partial address
3. Click "Complete Address" to see AI completion

### Format Standardization
1. Go to the "Standardize Format" tab
2. Enter an informal or messy address
3. Click "Standardize Format" to see cleaned version

## Example addresses

- **Complete**: C-704, Gayatri Shivam, Thakur Complex, Kandivali East, 400101
- **Partial**: C-704, Gayatri Shivam, Thakur Complex
- **Informal**: c704 gayatri shivam thakur complex kandivali e 400101

## Model Information

- **Base Model**: meta-llama/Llama-3.2-1B-Instruct
- **Parameters**: 1.24B
- **Model Size**: ~2.47GB
- **Max Context**: 131K tokens
- **License**: Apache 2.0

## Supported Components

The model can handle:
- Building names, localities, pincodes
- Cities, states, sub-localities  
- Road names, landmarks
- Various Indian address formats

## Performance Notes

⭐ **Entity Extraction**: Excellent performance - primary use case
⚠️ **Address Completion**: Limited training data - experimental feature  
⚠️ **Standardization**: Limited training data - experimental feature

**Recommendation**: Use this model primarily for **address component extraction** where it performs best.