Modalkit Examples¶

Comprehensive examples showcasing Modalkit's capabilities for deploying production-ready ML services on Modal.

🎯 Getting Started¶

Sentiment Analysis Service ¶

A complete sentiment analysis API demonstrating core Modalkit features.

What you'll learn: - Basic Modalkit setup and configuration - Text processing with transformers - Batch processing and error handling - Cloud storage integration - Production deployment patterns

Technologies: Transformers, PyTorch, Language Detection Difficulty: ⭐⭐☆☆☆

🚀 Advanced Examples¶

Async Processing with Queue Backends ¶

Flexible async processing with TaskIQ, SQS, or custom queue systems.

What you'll learn: - Optional queue backend integration - TaskIQ dependency injection - Custom queue implementations - Error-resilient async processing - Production queue setup patterns

Technologies: TaskIQ, Redis, SQS, RabbitMQ Difficulty: ⭐⭐⭐☆☆

🔗 Working Examples: See queue-patterns.md and taskiq-integration.md

Large Language Model Deployment ¶

Deploy and scale LLMs with GPU acceleration and intelligent batching.

What you'll learn: - GPU optimization for LLMs - Model loading from cloud storage - Intelligent batching strategies - Memory management techniques - Cost optimization approaches

Technologies: Transformers, PyTorch, HuggingFace Hub Difficulty: ⭐⭐⭐⭐☆

Computer Vision Pipeline ¶

Production computer vision service with classification and object detection.

What you'll learn: - Multi-task computer vision - Image preprocessing and augmentation - GPU acceleration for vision models - Object detection with YOLO - Performance monitoring

Technologies: OpenCV, Torchvision, Ultralytics, YOLO Difficulty: ⭐⭐⭐☆☆

Sophisticated AI service processing text, images, and audio simultaneously.

What you'll learn: - Cross-modal understanding - Vector embeddings generation - Semantic search capabilities - Advanced model orchestration - Complex input validation

Technologies: CLIP, Whisper, Sentence Transformers, Vector Databases Difficulty: ⭐⭐⭐⭐⭐

Real-Time Analytics Pipeline ¶

Stream processing and ML analytics for time-series data and events.

What you'll learn: - Stream processing patterns - Anomaly detection algorithms - Time-series forecasting - Event classification and alerting - Redis integration for real-time data

Technologies: Redis, Scikit-learn, Prophet, Pandas Difficulty: ⭐⭐⭐⭐☆

📊 Example Comparison¶

Example	Use Case	Complexity	GPU Required	Key Features
Sentiment Analysis	Text analysis API	Basic	Optional	Batch processing, cloud storage
Async Processing	Queue-based workflows	Intermediate	Optional	TaskIQ integration, dependency injection
LLM Deployment	Text generation	Advanced	Yes	GPU optimization, large models
Computer Vision	Image analysis	Intermediate	Recommended	Multi-task, object detection
Multi-Modal AI	Cross-modal understanding	Expert	Yes	Complex orchestration, embeddings
Real-Time Analytics	Stream processing	Advanced	No	Time-series, real-time processing

🛠️ Development Patterns¶

Each example demonstrates different Modalkit patterns:

Basic Patterns (Sentiment Analysis)¶

Simple inference pipeline
Configuration-driven deployment
Basic error handling
Standard authentication

Async Patterns (Async Processing)¶

Optional queue backend integration
TaskIQ dependency injection
Custom queue implementations
Error-resilient async processing

Performance Patterns (LLM, Computer Vision)¶

GPU acceleration techniques
Memory optimization
Intelligent batching
Model caching strategies

Complex data orchestration
Multiple model coordination
Real-time processing
Advanced monitoring

Production Patterns (All Examples)¶

Cloud storage integration
Comprehensive error handling
Monitoring and observability
Auto-scaling configuration

📁 Example Structure¶

Each example follows a consistent structure:

example-name/
├── README.md              # Tutorial walkthrough
├── app.py                 # Modal app definition
├── model.py               # ML model implementation
├── modalkit.yaml          # Configuration
├── requirements.txt       # Dependencies
├── utils/                 # Utility functions
│   ├── preprocessing.py
│   └── validation.py
└── tests/                 # Test files
    ├── test_model.py
    └── test_api.py

🚀 Quick Start¶

Choose an example based on your use case
Follow the tutorial step-by-step
Customize for your specific needs
Deploy to production

Prerequisites¶

Modal account and CLI installed
Python 3.9+
Basic familiarity with ML concepts

Installation¶

# Install Modalkit
pip install modalkit

# Install Modal CLI
pip install modal
modal setup

🎯 Use Case Guide¶

Text Processing¶

Simple text analysis → Sentiment Analysis
Text generation → LLM Deployment
Multi-language support → Multi-Modal AI

Async Processing¶

Queue-based workflows → Async Processing
TaskIQ integration → Async Processing
Custom queue systems → Async Processing

Computer Vision¶

Image classification → Computer Vision
Object detection → Computer Vision
Visual search → Multi-Modal AI

Audio Processing¶

Speech recognition → Multi-Modal AI
Audio classification → Multi-Modal AI

Analytics & Monitoring¶

Real-time metrics → Real-Time Analytics
Anomaly detection → Real-Time Analytics
Forecasting → Real-Time Analytics

📊 Performance Benchmarks¶

Latency Expectations¶

Sentiment Analysis: ~50-100ms per request
LLM Generation: ~1-5s per request (depends on length)
Computer Vision: ~100-300ms per image
Multi-Modal: ~200-500ms per request
Real-Time Analytics: ~10-50ms per event

Throughput Expectations¶

Text Processing: 100-1000 requests/second
Image Processing: 10-100 images/second
Audio Processing: 5-50 files/second
Analytics: 1000-10000 events/second

🔧 Configuration Templates¶

Development Environment¶

deployment_config:
  gpu: null
  concurrency_limit: 1
  container_idle_timeout: 300
  secure: false

Production Environment¶

deployment_config:
  gpu: "T4"  # or A10G/A100 for heavier workloads
  concurrency_limit: 10
  container_idle_timeout: 900
  secure: true
  retries: 3

High-Scale Environment¶

deployment_config:
  gpu: "A100"
  concurrency_limit: 50
  container_idle_timeout: 1800
  memory: 32768
  retries: 5

🤝 Contributing¶

Want to add your own example? See our Contributing Guidelines for:

Example requirements and standards
Code quality expectations
Documentation guidelines
Testing requirements

📚 Additional Resources¶

💡 Need Help?¶

Issues: GitHub Issues
Discussions: GitHub Discussions
Examples: Each example includes troubleshooting sections

These examples demonstrate Modalkit's power in deploying production-ready ML services with minimal boilerplate code. Start with the sentiment analysis example and progress to more complex use cases as you become familiar with the framework.