gpux 0.1.1

Docker-like GPU runtime for ML inference with universal GPU compatibility

GPUX - Docker-like GPU Runtime for ML Inference

GPUX provides universal GPU compatibility for ML inference workloads, allowing you to run the same model on any GPU without compatibility issues. Think Docker, but for ML models.

🚀 Features

• Universal GPU Support: Works on NVIDIA, AMD, Apple Silicon, and Intel GPUs
• Docker-like UX: Familiar commands and configuration files
• Zero Configuration: Automatically selects the best execution provider
• High Performance: Optimized ONNX Runtime backends
• Easy Deployment: Simple HTTP server for production use
• Cross-platform: Works on Windows, macOS, and Linux

📦 Installation

# Install with pip
pip install gpux

# Or install with uv (recommended)
uv add gpux

🎯 Quick Start

1. Create a gpux.yml

# gpux.yml
name: sentiment-analysis
version: 1.0.0

model:
  source: ./model.onnx
  format: onnx

inputs:
  text:
    type: string
    max_length: 512
    required: true

outputs:
  sentiment:
    type: float32
    shape: [1, 2]
    labels: [negative, positive]

runtime:
  gpu:
    memory: 2GB
    backend: auto

2. Build your model

gpux build .

3. Run inference

# Single inference
gpux run sentiment-analysis --input '{"text": "I love this product!"}'

# From file
gpux run sentiment-analysis --file input.json

# Benchmark
gpux run sentiment-analysis --benchmark --runs 1000

4. Start a server

gpux serve sentiment-analysis --port 8080

🛠️ Commands

Build

gpux build [PATH]                    # Build model from directory
gpux build . --provider cuda         # Build with specific GPU provider
gpux build . --no-optimize           # Build without optimization

Run

gpux run MODEL_NAME                  # Run inference
gpux run MODEL_NAME --input DATA     # Run with input data
gpux run MODEL_NAME --file FILE      # Run with input file
gpux run MODEL_NAME --benchmark      # Run benchmark

Serve

gpux serve MODEL_NAME                # Start HTTP server
gpux serve MODEL_NAME --port 9000    # Custom port
gpux serve MODEL_NAME --workers 4    # Multiple workers

Inspect

gpux inspect MODEL_NAME              # Inspect model
gpux inspect --model model.onnx      # Inspect model file directly
gpux inspect --json                  # JSON output

🔧 Configuration

gpux.yml Format

name: model-name
version: 1.0.0
description: "Model description"

model:
  source: ./model.onnx
  format: onnx

inputs:
  input_name:
    type: float32
    shape: [1, 10]
    required: true
    description: "Input description"

outputs:
  output_name:
    type: float32
    shape: [1, 2]
    labels: [class1, class2]
    description: "Output description"

runtime:
  gpu:
    memory: 2GB
    backend: auto  # auto, cuda, coreml, rocm, vulkan, metal, dx12
  batch_size: 1
  timeout: 30

serving:
  port: 8080
  host: 0.0.0.0
  batch_size: 1
  timeout: 5

preprocessing:
  tokenizer: bert-base-uncased
  max_length: 512
  resize: [224, 224]
  normalize: imagenet

🎯 Supported Platforms

Platform	GPU	Provider	Status
NVIDIA	CUDA	CUDAExecutionProvider	✅
NVIDIA	TensorRT	TensorrtExecutionProvider	✅
AMD	ROCm	ROCmExecutionProvider	✅
Apple	Metal	CoreMLExecutionProvider	✅
Intel	OpenVINO	OpenVINOExecutionProvider	✅
Windows	DirectML	DirectMLExecutionProvider	✅
Universal	CPU	CPUExecutionProvider	✅

🚀 Performance

GPUX automatically selects the best execution provider for your hardware:

• Apple Silicon: CoreML (optimized for M1/M2/M3)
• NVIDIA: TensorRT > CUDA (best performance)
• AMD: ROCm (ROCm acceleration)
• Intel: OpenVINO (Intel optimization)
• Windows: DirectML (Windows GPU acceleration)
• Fallback: CPU (universal compatibility)

📚 Examples

Check out the examples/ directory for complete examples:

• Sentiment Analysis - BERT-based text classification
• Image Classification - ResNet-50 for ImageNet

🔌 API Reference

Python API

from gpux import GPUXRuntime

# Initialize runtime
runtime = GPUXRuntime(model_path="model.onnx")

# Run inference
results = runtime.infer({"input": data})

# Benchmark
metrics = runtime.benchmark(data, num_runs=100)

# Get model info
info = runtime.get_model_info()

HTTP API

When serving a model, GPUX provides a REST API:

# Health check
GET /health

# Model information
GET /info

# Run inference
POST /predict
Content-Type: application/json

{
  "input": "your data here"
}

🧪 Testing

# Run tests
pytest

# Run with coverage
pytest --cov=src/gpux

# Run specific test
pytest tests/test_runtime.py

🤝 Contributing

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Add tests
5. Submit a pull request

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

• ONNX Runtime for the excellent ML runtime
• FastAPI for the web framework
• Typer for the CLI framework
• Rich for beautiful terminal output

📞 Support

• 📖 Documentation
• 🐛 Issues
• 💬 Discussions
• 📧 Email