MLX Models for Apple Silicon

🚀 Production-ready MLX implementations of popular AI models, optimized for Apple Silicon (M1/M2/M3/M4)

🌟 Available Models

1. PaddleOCR-VL-MLX

World's first MLX-native OCR model

• 🔗 Model on Hugging Face
• 📦 Size: ~2GB
• ⚡ Speed: 2-3s per image on M4 Max
• 🎯 Use case: Document digitization, receipt processing, academic paper OCR
• 🌐 Web UI Available - Beautiful browser-based interface!

2. Hunyuan-MT-Chimera-7B-MLX-Q8

8-bit quantized multilingual translation model

• 🔗 Model on Hugging Face
• 📦 Size: 4.2GB (70% smaller than original)
• ⚡ Speed: 25 tokens/s on M4 Max
• 🌍 Languages: 200+
• 🎯 Use case: Document translation, real-time translation, multilingual chat

🚀 Quick Start

Installation

# Install MLX and dependencies
pip install mlx mlx-lm transformers pillow

# Clone this repo for examples
git clone https://github.com/biabia-55/mlx-models.git
cd mlx-models

OCR Example

from transformers import AutoTokenizer
from PIL import Image

# Load model
tokenizer = AutoTokenizer.from_pretrained(
    "biabia-55/PaddleOCR-VL-MLX",
    trust_remote_code=True
)

# Process image
image = Image.open("document.jpg")
# See paddleocr-vl-mlx/examples/ for complete code

Translation Example

from mlx_lm import load, generate

# Load model
model, tokenizer = load("biabia-55/Hunyuan-MT-Chimera-7B-MLX-Q8")

# Translate
prompt = "Translate to French: Hello, world!"
result = generate(model, tokenizer, prompt=prompt, max_tokens=256)
print(result)

📊 Performance Benchmarks

Tested on M4 Max (128GB RAM):

Model	Task	Speed	Memory	vs PyTorch
PaddleOCR-VL-MLX	OCR	2-3s/image	~4GB	2.5x faster
Hunyuan-MT-MLX-Q8	Translation	25 tok/s	~8GB	1.67x faster

📁 Repository Structure

mlx-models/
├── paddleocr-vl-mlx/          # PaddleOCR-VL examples and tools
│   ├── examples/              # Usage examples
│   ├── conversion/            # PyTorch → MLX conversion scripts
│   └── README.md
├── hunyuan-mt-mlx/            # Hunyuan-MT examples and tools
│   ├── examples/              # Usage examples
│   ├── quantization/          # Quantization scripts
│   └── README.md
├── docs/                      # Documentation
│   ├── performance_benchmarks.md
│   ├── conversion_guide.md
│   └── faq.md
└── README.md                  # This file

🎯 Use Cases

Document Processing Pipeline

# See paddleocr-vl-mlx/examples/document_pipeline.py
from paddleocr_mlx import OCRPipeline

pipeline = OCRPipeline()
result = pipeline.process("document.pdf")
# Preserves formatting, tables, and layout

Real-time Translation

# See hunyuan-mt-mlx/examples/streaming_translation.py
from hunyuan_mlx import StreamingTranslator

translator = StreamingTranslator()
for chunk in translator.translate_stream("Long text..."):
    print(chunk, end='', flush=True)

Batch Processing

# See examples for batch processing multiple files
# Optimized for throughput on Apple Silicon

🛠️ Development

Converting Your Own Models

See docs/conversion_guide.md for detailed instructions on converting PyTorch models to MLX.

Running Tests

python -m pytest tests/

📚 Documentation

• Performance Benchmarks
• Conversion Guide
• FAQ
• API Reference

🤝 Contributing

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

1. Fork the repository
2. Create your feature branch (git checkout -b feature/AmazingFeature)
3. Commit your changes (git commit -m 'Add some AmazingFeature')
4. Push to the branch (git push origin feature/AmazingFeature)
5. Open a Pull Request

📄 License

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

🙏 Acknowledgments

• PaddlePaddle Team for PaddleOCR-VL
• Tencent Hunyuan Team for Hunyuan-MT
• Apple ML Research for the MLX framework
• Hugging Face for model hosting

📮 Contact

• GitHub: @biabia-55
• Hugging Face: @gamhtoi
• Email: wheiwalee@gmail.com

Note: Models are hosted on Hugging Face under gamhtoi, while code and documentation are on GitHub under biabia-55. Both accounts belong to the same author.

⭐ Star History

If you find this project useful, please consider giving it a star!

---

Made with ❤️ for the Apple Silicon community