Add enhanced Docker setup with management tools and CUDA 13.0 support

[loserbcc]

Summary

This PR adds a comprehensive Docker containerization solution for ACE-Step, optimized for modern NVIDIA GPUs and production deployment.

Features

🐳 Docker Infrastructure

• Optimized Dockerfile (Dockerfile.scorpy) with CUDA 13.0 support for RTX 5090 and newer GPUs
• Docker Compose configuration with full GPU support and health checks
• Persistent volumes for checkpoints, outputs, and logs
• CPU offload mode enabled (~0.4GB VRAM usage)

🛠️ Management Tools

• Quick launcher (ace-docker) - Single command to start/stop/manage
• Full management script (docker-manage.sh) - Complete lifecycle management
• Helper scripts for different quality presets
• Comprehensive documentation (DOCKER.md)

⚡ Performance Optimizations

• BFloat16 precision for faster inference
• Overlapped decode for better throughput
• Tested on RTX 5090 with excellent performance

Usage

# Ultra-simple launch
./ace-docker

# Or use full management interface
./docker-manage.sh start   # Start container
./docker-manage.sh logs    # View logs
./docker-manage.sh status  # Check status
./docker-manage.sh rebuild # Full rebuild

Access the web interface at: http://localhost:7866

Files Added

• .dockerignore - Build optimization
• Dockerfile.scorpy - CUDA 13.0 optimized build
• docker-compose.scorpy.yml - Container orchestration
• docker-manage.sh - Management script
• ace-docker - Quick launcher
• DOCKER.md - Complete documentation
• Helper scripts for quality presets

Testing

This setup has been tested and verified working on:

• ✅ RTX 5090 with CUDA 13.0
• ✅ Arch Linux (should work on any Linux with Docker)
• ✅ Production deployment with persistent volumes

Notes

• Uses .scorpy suffix to avoid conflicts with existing Docker files
• Complements rather than replaces existing Docker setup
• All volumes are persistent and survive container restarts
• Includes health checks and proper logging

Documentation

Complete setup instructions, troubleshooting, and advanced usage examples are included in DOCKER.md.

---

🤖 Generated with Claude Code

[tailoralm]

thank you so much! I have an RTX 5070 and you saved me