search

ml-explore / mlx-examples

Examples in the MLX framework
MIT License
5.92k stars 838 forks source link

Add Support for Full Model Fine-Tuning #831

Closed Goekdeniz-Guelmez closed 2 months ago

Goekdeniz-Guelmez commented 3 months ago

This pull request introduces the following enhancements to the mlx-lm package:

Full Model Fine-Tuning:

Added the --train-full flag to enable fine-tuning of the full model weights. Also updated the LORA.md and ACKNOWLEDGMENTS.md files.

Works as intended, tried it with tiny-random-Gemma, OpenELM-450M, and OpenELM-270M.

awni commented 3 months ago

This looks nice and simple though how do you actually load the fine-tuned model? I don't think everything you need to load it is saved in the adapter_path?