Request for Apple Silicon (M-series) support and MPS backend compatibility

[BroccoliFin]

Description

I'm interested in running FlashLabs-Chroma on a Mac with Apple Silicon (e.g., M1/M2/M3 chips). However, the current installation relies on CUDA, which isn't supported on Apple hardware. PyTorch has MPS backend for acceleration on Mac GPUs, but the README and requirements don't mention it.

Steps to Reproduce / Attempted Installation

1. Cloned the repo and created conda env with Python 3.11.
2. Tried pip install -r requirements.txt — fails due to CUDA dependency (no CUDA on Mac).
3. Alternative: Installed PyTorch with MPS support (pip install torch), but model code assumes CUDA (e.g., device mapping).

Expected Behavior

• Instructions for MPS/CPU installation.
• Code adaptations to use device='mps' or fallback to CPU.
• Testing on Apple Silicon for real-time performance.

Environment

• OS: macOS [your version, e.g., Ventura 13.x]
• Hardware: MacBook Pro M2, 16GB RAM
• Python: 3.11

Would appreciate any guidance or plans to add support! This would make the project more accessible.

[ZhenghuaBao]

Hi @BroccoliFin,

Thank you for the detailed issue report! You're correct that the current setup assumes CUDA availability, which isn't compatible with Apple Silicon.

Current status:

• The project is optimized for GPUs with CUDA
• MPS support for Apple Silicon isn't implemented yet

Regarding your hardware:

• 16GB RAM may be insufficient for optimal performance - we recommend at least 30GB for consistent real-time inference

We're aware of the need for broader hardware compatibility and are exploring options for Mac systems. We'll update the community if we make progress on MPS support or CPU-optimized configurations.