Sinhala Offline Chatbot

Fully offline Sinhala chatbot Local inference with OLLAMA Streamlit-based interactive UI Prompt engineering + rewrite mechanism

Features

• Sinhala input and Sinhala output
• Fully offline local inference
• Streamlit chat interface
• Session-based chat history
• Model comparison panel
• Testing panel with predefined prompts
• Sinhala rewrite mechanism
• Chat export

Technologies

• Python
• Streamlit
• OLLAMA
• Requests

Project Structure

• app.py – main Streamlit app
• chatbot.py – model interaction and response logic
• prompts.py – system prompts and few-shot examples
• ui.py – UI rendering and styling
• utils.py – helper functions
• requirements.txt – dependencies

How to Run

1. Install dependencies:

   pip install -r requirements.txt

2. Start OLLAMA:

   ollama serve

3. Pull required models:

   ollama pull Tharusha_Dilhara_Jayadeera/singemma
   ollama pull llama3:latest

4. Run the app:

   streamlit run app.py
   

Evaluation Summary

• SinGemma performed better overall in this implementation
• Llama 3 produced fewer usable outputs in the recorded test set
• Rewrite mechanism improved mixed-language output quality

Screenshots

Main Chat Interface

Sinhala Interaction Example

Model Comparison Panel

Testing Panel