SUMMARY.md

Qwen Multi-label Text Classifier

Overview

A multi-label text classifier based on Qwen-1.5B, fine-tuned for coding exercise classification. Supports:

• Local CPU/GPU inference
• Hugging Face API deployment
• Batch evaluation
• REST API via FastAPI
• Docker deployment

Features

• 9 Label Classification: Predicts multiple tags per text
• CLI Interface: Run predictions/evaluations from terminal
• Dual Backend: Choose between local or HF inference
• GPU Optimized: CUDA support via Docker

Installation

git clone https://github.com/your-username/qwen-classifier
cd qwen-classifier
python3 -m venv .venv
source .venv/bin/activate
pip install -e .

Usage

CLI Prediction

# Local inference
qwen-clf predict "Your coding exercise text" --backend local

# HF Space inference 
qwen-clf predict "Your text" --backend hf --hf-token YOUR_TOKEN

Batch Evaluation

qwen-clf evaluate dataset.zip --backend local

API Server

uvicorn app:app --host 0.0.0.0 --port 7860

API Endpoints

Endpoint	Method	Description
/	GET	Documentation
/predict	POST	Single text prediction
/evaluate	POST	Batch evaluation (ZIP)
/health	GET	Service status

Docker Deployment

# Build with GPU support
docker build -t qwen-classifier .

# Run container
docker run -p 7860:7860 --gpus all qwen-classifier

Project Structure

.
├── app.py               # FastAPI entry point
├── Dockerfile           # GPU-optimized container setup
├── qwen_classifier/     # Core package
│   ├── cli.py          # Command line interface
│   ├── model.py        # Qwen classifier implementation  
│   ├── predict.py      # Inference logic
│   └── evaluate.py     # Batch evaluation
└── requirements.txt    # Python dependencies

Configuration

Edit qwen_classifier/config.py to set:

• TAG_NAMES: List of 9 classification tags
• HF_REPO: Default Hugging Face model repo
• DEVICE: Auto-detected CUDA/CPU

Hugging Face Space

Live demo:

License

Apache 2.0 © Keivan Razban