Zum Inhalt

How You Implement Own Training (MacOS)

We are currently working on this guide and collecting more real-world usage experience from users to provide a clearer, more effective walkthrough.

However, we want to emphasize that it is technically fully possible to implement and run your own AI training code inside DeepExtension on MacOS.

What You Can Do Today

We recommend reviewing How We Implemented MLX-Demo (MacOS), which demonstrates a successful integration between DeepExtension and the mlx_lm.lora training interface.

If your training framework is implemented in a similar style to mlx_lm.lora, it can be integrated into DeepExtension with minimal adaptation.

What Comes Next

We are actively working on:

  • Testing more training frameworks on MacOS (especially MLX-based ones)
  • Providing step-by-step templates and working demo scripts
  • Aligning the logging and output saving logic with Mac-specific runtime constraints

Your feedback and contributions are very welcome. A little more patience from our Mac users will help us deliver a well-rounded and robust developer experience.

DeepExtension — enabling Mac-native AI development without compromise.