axolotl
axolotl copied to clipboard
MLX Support
Hi, It would be great to have MLX support in Axolotl. MLX has been shown to be able to quickly and efficiently finetune many LLMs, including 7B LLMs on consumer hardware. Thank you!
(edit: update)
These will be helpful
https://github.com/philipturner/metal-flash-attention
https://github.com/ml-explore/mlx/issues/129
Since #1264 is merged, should I close this issue or are there plans to implement MLX support in the future (since MLX can be much faster)?