Notes on the Apple MLX machine learning framework

Apple MLX for AI/Large Language Models—Day One

Rendered markdown LinkedIn version Jupyter notebook

Converting models from Hugging Face to MLX format, and sharing

Rendered markdown LinkedIn version Jupyter notebook

Retrieval augmentation with MLX: A bag full of RAG, part 1

Rendered markdown LinkedIn version Jupyter notebook

Retrieval augmentation with MLX: A bag full of RAG, part 2

Rendered markdown LinkedIn version Jupyter notebook

More (up to date) MLX resources

• MLX home page
• Hugging Face MLX community
• Using MLX at Hugging Face
• MLX Text-completion Finetuning Notebook
• MLX Tuning Fork—Framework for parameterized large language model (Q)LoRa fine-tuning using mlx, mlx_lm, and OgbujiPT. Architecture for systematic running of easily parameterized fine-tunes

A few general notes

• For the many chat formats already charted out in llama.cpp, see the @register_chat_format decorated functions in llama_chat_format.py

To do, or figure out

• any grammar/ebnf support a la llama.cpp?
• Alternate LLM sampling methods
• Steering vectors

Syncing articles to notebooks

Use Jupytext to convert the .md articles to .ipynb notebooks:

jupytext --to ipynb 2024/MLX-day-one.md

May have to convert cells using plain pip to use %pip instead. It also doesn't seem to check the format metadata, so you might need to convert non-Python cells back to Markdown by hand.

License

Shield:

This work is licensed under a Creative Commons Attribution 4.0 International License.

See also: https://github.com/santisoler/cc-licenses?tab=readme-ov-file#cc-attribution-40-international