README.md

metadata

title: LLM Data Analyzer
emoji: 📊
colorFrom: blue
colorTo: indigo
sdk: docker
sdk_version: latest
app_file: app.py
pinned: false

📊 LLM Data Analyzer

An AI-powered tool for analyzing data and having conversations with an intelligent assistant powered by Llama 2.

Features

• 📤 Upload & Analyze: Upload CSV or Excel files and get instant analysis
• 💬 Chat: Have conversations with Llama 2 AI assistant
• 📊 Data Statistics: View comprehensive data summaries and insights
• 🚀 Fast: Runs on free Hugging Face CPU tier

How to Use

1. Upload Data - Start by uploading a CSV or Excel file
2. Preview - Review your data and statistics
3. Ask Questions - Get AI-powered analysis and insights
4. Chat - Have follow-up conversations with the AI

Technology Stack

• Model: Llama 2 7B (quantized to 4-bit)
• Framework: Streamlit
• Inference Engine: Llama.cpp
• Hosting: Hugging Face Spaces
• Language: Python 3.10+

Performance

Metric	Value
Speed	~5-10 tokens/second (free CPU)
Model Size	4GB (quantized)
Context Window	2048 tokens
First Load	~30 seconds (model download)
Subsequent Responses	~5-15 seconds
Hardware	Free Hugging Face CPU

Local Development (Faster)

For faster local development with GPU acceleration on Apple Silicon Mac:

# Clone the repository
git clone https://github.com/Arif-Badhon/LLM-Data-Analyzer
cd LLM-Data-Analyzer

# Switch to huggingface-deployment branch
git checkout huggingface-deployment

# Install dependencies
pip install -r requirements.txt

# Run with MLX (Apple Silicon GPU - ~70 tokens/second)
streamlit run app.py

Deployment Options

Option 1: Hugging Face Space (Free)

• CPU-based inference
• Speed: 5-10 tokens/second
• Cost: Free

Option 2: Local with MLX (Fastest)

• GPU-accelerated on Apple Silicon
• Speed: 70+ tokens/second
• Cost: Free (uses your Mac)

Option 3: Hugging Face PRO (Fast)

• GPU-accelerated inference
• Speed: 50+ tokens/second
• Cost: $9/month

Getting Started

Quick Start (3 minutes)

# 1. Install Python 3.10+
# 2. Clone repo
git clone https://github.com/Arif-Badhon/LLM-Data-Analyzer
cd LLM-Data-Analyzer

# 3. Install dependencies
pip install -r requirements.txt

# 4. Run Streamlit app
streamlit run app.py

With Docker (Local Development)

# Make sure Docker Desktop is running
docker-compose up --build

# Access at http://localhost:8501

Troubleshooting

"Model download failed"

• Check internet connection
• HF Spaces need internet to download models from Hugging Face Hub
• Wait and refresh the page

"App takes too long to load"

• Normal on first request (10-30 seconds)
• Model is being downloaded and cached
• Subsequent requests are much faster

"Out of memory"

• Free tier CPU is limited
• Unlikely with quantized 4GB model
• If it happens, upgrade to HF PRO

"Slow responses"

• Free tier CPU is slower than GPU
• Expected: 5-10 tokens/second
• For faster responses: use local MLX (70 t/s) or upgrade HF tier

Technologies Used

• Python - Core language
• Streamlit - Web UI framework
• Llama 2 - Large language model
• Llama.cpp - CPU inference
• MLX - Apple Silicon GPU inference
• Pandas - Data processing
• Docker - Containerization
• Hugging Face Hub - Model hosting

License

MIT License

Author

Arif Badhon

Support

If you encounter any issues:

1. Check the Troubleshooting section above
2. Review Hugging Face Spaces Docs
3. Open an issue on GitHub

---

Happy analyzing! 🚀