Mobile-O-0.5B-iOS

Optimized MLX & CoreML Components for On-Device Deployment

📌 Overview

This repository contains the optimized MLX and CoreML model components of Mobile-O-0.5B for native iOS deployment. These components power the Mobile-O iOS app, enabling fully on-device multimodal understanding and image generation with no cloud dependency.

📱 On-Device Performance

Spec	Detail
⚡ Image Generation	~3 seconds
👁️ Visual Understanding	~0.4 seconds
💾 Memory Footprint	< 2GB
📱 Compatible Devices	iPhone (A17+ / M-series)
🔒 Cloud Dependency	None — fully on-device

📦 Contents

This repo includes optimized model components in both MLX and CoreML formats:

Component	Format	Description
VLM	MLX / CoreML	FastVLM-0.5B (FastViT + Qwen2-0.5B)
Diffusion Decoder	MLX / CoreML	SANA-600M-512 (Linear DiT + VAE)
MCP	MLX / CoreML	Mobile Conditioning Projector (~2.4M params)

🚀 Usage

With the iOS App

1. Clone the Mobile-O repo
2. Navigate to the Mobile-O-App/ directory
3. Download this model repo into the app's model directory
4. Build and run in Xcode

git clone https://github.com/Amshaker/Mobile-O.git
cd Mobile-O/Mobile-O-App

Refer to the Mobile-O-App README for detailed setup instructions.

Download Models

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="Amshaker/Mobile-O-0.5B-iOS",
    repo_type="model",
    local_dir="ios_models"
)

🔗 Related Resources

Resource	Link
🤗 Mobile-O-0.5B	PyTorch Model
🤗 Mobile-O-1.5B	PyTorch Model
📱 iOS App Source Code	Mobile-O-App
🤗 Training Datasets	Collection

📄 Citation

@article{shaker2026mobileo,
  title={Mobile-O: Unified Multimodal Understanding and Generation on Mobile Device},
  author={Shaker, Abdelrahman and Heakl, Ahmed and Muhammad, Jaseel and Thawkar, Ritesh and Thawakar, Omkar and Li, Senmao and Cholakkal, Hisham and Reid, Ian and Xing, Eric P. and Khan, Salman and Khan, Fahad Shahbaz},
  journal={arXiv preprint arXiv:XXXX.XXXXX},
  year={2026}
}

⚖️ License

Released under CC BY-NC 4.0. For research purposes only.