Docs/MLOPS_README.md

🤖 MLOps Training Platform

A beginner-friendly, multilingual MLOps platform for training text classification models. Built with Streamlit and PyTorch, this platform enables users to train, evaluate, and deploy machine learning models for content detection across multiple languages.

✨ Features

🌍 Multilingual Support

• English 🇬🇧 - Standard NLP preprocessing
• Chinese (中文) 🇨🇳 - Jieba word segmentation
• Khmer (ភាសាខ្មែរ) 🇰🇭 - Unicode normalization for Khmer script

🎯 Key Capabilities

• Easy Data Upload - Upload CSV files with your training data
• Interactive Configuration - Adjust hyperparameters with sliders and dropdowns
• Real-time Training Monitoring - Watch progress and metrics as your model trains
• Comprehensive Evaluation - View confusion matrices, accuracy, precision, recall, F1
• Model Export - Download trained models for deployment

🧠 Supported Model Architectures

Model	Languages	Description
mBERT	EN, ZH, KM	Multilingual BERT supporting 104 languages
XLM-RoBERTa	EN, ZH, KM	Cross-lingual model with excellent performance
DistilBERT	EN, ZH, KM	Lightweight, faster training
RoBERTa	EN	Optimized for English text

🚀 Quick Start

Prerequisites

• Python 3.8 or higher
• pip package manager
• 4GB+ RAM recommended

Installation

1. Clone the repository

git clone <repository-url>
cd "90. Content Detection Template"

2. Create a virtual environment (recommended)

python -m venv venv

# Windows
venv\Scripts\activate

# Linux/Mac
source venv/bin/activate

3. Install dependencies

pip install -r requirements.txt

4. Run the Streamlit app

streamlit run streamlit_app.py

5. Open your browser Navigate to http://localhost:8501

📖 User Guide

Step 1: Upload Your Data

Your CSV file should have at least two columns:

• text - The text content to classify
• label - Binary labels (0 or 1)

Example CSV format:

text,label
"This is a legitimate message.",0
"URGENT: Click here to claim your prize!",1
"Meeting scheduled for tomorrow at 3pm.",0
"Your account has been compromised!",1

Step 2: Configure Training

1. Select Target Language - Choose the language of your data
2. Choose Model Architecture - Select based on your needs:
   • Use DistilBERT for faster training on CPU
   • Use mBERT or XLM-RoBERTa for best multilingual performance
3. Set Hyperparameters:
   • Learning Rate: Start with 2e-5 (default)
   • Epochs: 3-5 is usually sufficient
   • Batch Size: Use 8-16 for CPU, larger for GPU

Step 3: Train Your Model

1. Click Start Training
2. Monitor progress in real-time
3. View training metrics as they update

Step 4: Evaluate & Download

1. Review final metrics (accuracy, precision, recall, F1)
2. Test the model with new text samples
3. Download the trained model as a ZIP file

📁 Project Structure

├── streamlit_app.py          # Main Streamlit application
├── mlops/                    # MLOps backend modules
│   ├── __init__.py
│   ├── config.py             # Configuration classes
│   ├── preprocessor.py       # Language-specific preprocessing
│   ├── trainer.py            # Model training logic
│   └── evaluator.py          # Model evaluation & visualization
├── app/                      # FastAPI backend (original)
│   └── main.py
├── models/                   # Pre-trained models
├── trained_models/           # Output directory for trained models
├── requirements.txt          # Python dependencies
└── README.md                 # This file

🔧 Configuration Options

Training Parameters

Parameter	Default	Description
Learning Rate	2e-5	Model learning rate
Epochs	3	Number of training epochs
Batch Size	16	Training batch size
Max Length	256	Maximum token sequence length
Train Split	80%	Percentage for training
Validation Split	10%	Percentage for validation
Test Split	10%	Percentage for testing

Advanced Options

Parameter	Default	Description
Warmup Ratio	0.1	LR warmup fraction
Weight Decay	0.01	L2 regularization
Random Seed	42	For reproducibility

🌐 API Integration

The platform also includes a FastAPI backend for programmatic access:

import requests

# Make a prediction
response = requests.post(
    "http://localhost:8000/predict",
    json={"text": "Your text here"}
)
print(response.json())

🐳 Docker Support

# Build the image
docker build -t mlops-platform .

# Run the container
docker run -p 8501:8501 mlops-platform

📊 Sample Datasets

Load sample data directly from the UI by clicking the Sample button in the sidebar. Sample data is available for:

• English phishing detection
• Chinese phishing detection (中文钓鱼检测)
• Khmer content classification (ការចាត់ថ្នាក់មាតិកាខ្មែរ)

🛠️ Troubleshooting

Common Issues

Out of Memory Error

• Reduce batch size
• Use a smaller model (DistilBERT)
• Reduce max sequence length

Slow Training

• Ensure you're using GPU if available
• Reduce number of epochs
• Use a smaller dataset for testing

Model Loading Errors

• Ensure internet connection for downloading models
• Check available disk space
• Try a different model architecture

📝 License

This project is licensed under the MIT License.

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📧 Support

For issues and feature requests, please open a GitHub issue.

---

Built with ❤️ using Streamlit & PyTorch

Supports: 🇬🇧 English | 🇨🇳 中文 | 🇰🇭 ភាសាខ្មែរ