Boyun7/vision-language-chat-313831025

🤖 Vision-Language Chat Assistant

🔗 Live Demo: https://huggingface.co/spaces/Boyun7/vision-language-chat-313831025

Course: Data Science - Assignment 3 Student ID: 313831025 Model: SmolVLM-Instruct (256M parameters)

📋 Project Overview

A sophisticated web-based conversational AI application that enables natural, multi-turn dialogue about images using SmolVLM-Instruct, an ultra-lightweight Vision-Language model optimized for CPU inference.

✨ Key Features

• —🗣️ Multi-turn Conversation - Engage in natural back-and-forth dialogue about images
• —🎯 Context-Aware - Maintains conversation history for coherent responses
• —🔍 Detailed Analysis - Deep understanding of image content, objects, and scenes
• —⚡ CPU Optimized - Fast inference with only 256M parameters
• —💾 Export Capability - Save conversation history for future reference
• —📸 Auto-download Examples - Automatically downloads free test images from Unsplash

🏗️ Architecture

Technology Stack

Frontend:  Gradio 5.49.1 (Interactive Web UI)
Backend:   Python 3.10
Model:     SmolVLM-Instruct (256M parameters)
Framework: PyTorch + Transformers
Deployment: HuggingFace Spaces (CPU)

Model Details

SmolVLM-Instruct by HuggingFace

• —Size: 256M parameters (ultra-lightweight!)
• —Optimization: Designed for CPU inference
• —Speed: Fast response times even on CPU
• —Capabilities:
• —Image understanding
• —Multi-turn conversation
• —Object detection and description
• —Scene analysis

💡 Usage Guide

Basic Usage

1. 1.Upload an Image
2. 2.Click the image upload area
3. 3.Select an image from your device
4. 4.Or try one of the auto-downloaded example images

1. 1.Ask Questions
2. 2.Type your question in the text box
3. 3.Press Enter or click "Send 📤"
4. 4.View the AI's response in the chat area

1. 1.Continue the Conversation
2. 2.Ask follow-up questions
3. 3.Reference previous responses
4. 4.Explore different aspects of the image

1. 1.Manage Your Session
2. 2.Clear Chat: Reset the conversation
3. 3.Export Chat: Save conversation history

Example Questions

General Understanding:

• —"What objects can you see in this image?"
• —"Describe the scene in detail."
• —"What is happening in this image?"

Specific Details:

• —"What colors are dominant in this image?"
• —"How many people are in this image?"
• —"What's the weather like in this scene?"

🚀 Running Locally

If you want to run this locally:

# Clone the repository
git clone https://huggingface.co/spaces/Boyun7/vision-language-chat-313831025
cd vision-language-chat-313831025

# Install dependencies
pip install -r requirements.txt

# Run the application
python app.py

The app will:

1. 1.Download 3 example images (if not present)
2. 2.Load SmolVLM-Instruct model (~500MB download first time)
3. 3.Launch the web interface at http://localhost:7860

📊 Performance

Model Specs:

• —Parameters: 256M (ultra-lightweight)
• —Download Size: ~500MB
• —Memory Usage: ~2GB RAM
• —Inference Speed: 1-3 seconds per query (CPU)
• —First Load: ~30 seconds

Hardware Requirements:

• —CPU: Any modern CPU (no GPU needed!)
• —RAM: 4GB minimum, 8GB recommended
• —Storage: 1GB for model cache

🎓 Assignment Details

Course: Data Science Assignment: HW3 - Vision Language AI Student ID: 313831025 Submission Date: 2025-11-09

Features Implemented:

• —✅ Web UI with Gradio
• —✅ Vision-Language AI (SmolVLM-Instruct)
• —✅ Multi-turn conversation
• —✅ Auto-download example images
• —✅ Context-aware responses
• —✅ Chat export functionality
• —✅ CPU optimized for accessibility

📝 Technical Notes

Why SmolVLM-Instruct?

• —Ultra-lightweight (256M parameters)
• —Optimized for CPU inference
• —Fast response times
• —Good balance of size and capability
• —Perfect for educational demos
• —Accessible without GPU

Auto-download Images:

• —Images from Unsplash.com (free license)
• —Downloads on first run
• —Cached for subsequent runs

📧 Contact

Student ID: 313831025 Project: Vision-Language Chat Assistant Model: SmolVLM-Instruct by HuggingFace

📄 License

• —Code: Apache 2.0
• —SmolVLM-Instruct: Apache 2.0 (HuggingFace)
• —Example Images: Unsplash License

Enjoy exploring images with AI! 🚀