MLX Guru is here to help! 😃

[sck-at-ucy]

Happy to share "MLX Guru” 🚀.

It knows the entire MLX documentation (and will be updated as the framework evolves) and can help with your MLX code development. It has the big advantage that at the same it can generate high quality code in Python in general, and in Torch in particular, and understands the nuances of the differences between MLX and Torch so it can handle situations where the two meet each other in your code. Not perfect but I think it speeds up development especially if you are new to MLX.

This is shared as a big thank you to @awni and @angeloskath and the rest of the MLX team.

It is, and will be free, but it requires a ChatGPT Plus subscription at OpenAI.

Available here: https://chat.openai.com/g/g-7NeyFkq2e-mlx-guru

Let me know if you have any suggestions for improvements and of your experience using it if you decide to try it out.

[aminekhelif]

Hi,
I need to use a Python library called PyPose, which is built with PyTorch. However, PyTorch's Metal Performance Shaders (MPS) support is not yet implemented in PyPose. This is because some PyTorch methods that PyPose relies on do not exist in MPS. I have asked the PyPose developers if they plan to update the library to support Apple hardware, but they indicated that while it is a good idea, they do not plan to implement this in the near future.

As a result, I decided to convert this library to use MLX, but the new chatbots do not support either MLX or PyPose. The only code assistant I have found is you, GPT. So, I have come to you to ask if it is possible to have a conversation about potentially porting the knowledge from your GPT model into a local model file or chatbot.

Thank you in advance.

[demhademha]

Now that we have Swift API, would it be possible to update this bot to have support with that also?
Thank you

[sck-at-ucy]

That is a good idea.

Currently, there seems to be some sort of bug that prevents updates to customs GPTs (at least to MLX Guru), have been unable to update to 0.19.0., so we might have to re-configure from scratch.

Wonder it is better to have a separate version "MLX Guru Swift". Let me experiment a bit with combined and separate versions to see which method yields the best results.

[sck-at-ucy]

There is now a first version of "MLX-Swift Guru"

https://chatgpt.com/g/g-64bQ7zIwz-mlx-swift-guru

If you try it and have any comments, let me know to try and improve.

[johnmai-dev]

Can it support porting Python to Swift, reducing our workload of migrating the model to MLX Swift?

MLX Python: https://github.com/ml-explore/mlx-examples/tree/main/llms/mlx_lm/models
MLX Swift Models: https://github.com/ml-explore/mlx-swift-examples/tree/main/Libraries/LLM/Models