Porting PyTorch / CUDA to Apple MLX
What this skill covers
This skill captures the workflow and pitfalls accumulated across ~10 production MLX ports (LTX-2, CogVideoX-Fun, Matrix-Game, VOID, Fish S2 Pro, Mistral Small, Qwen Image, Hunyuan3D-2.1, …). It is for inference-only ports on Apple Silicon using the mlx-forge (conversion) + mlx-arsenal (reusable ops) + *-mlx fork convention.
Scope boundary — delegate:
- For writing or updating an *mlx-forge conversion recipe specific
[Description truncada. Veja o README completo no GitHub.]