Local AI powered commit messages! For apple silicon devices, because we are using mlx. Similar flow for traditional transformers model for non apple silicon devices, hook is the same.
Get the huggingface-cli via brew:
brew install huggingface-cli
Add your account token by logging in and generating one, then run and enter it:
huggingface-cli login
Download an mlx-community model from huggingface like:
huggingface-cli download mlx-community/DeepSeek-R1-Distill-Qwen-32B-MLX-4Bit
Create a new pyenv somewhere like:
python -m venv <venv_name>
This will create a directory named <venv_name>, after you want to activate it:
source <venv_name>/bin/activate
Install mlx_lm:
pip intall mlx_lm
Serve it over an open ai compat server, using mlx:
mlx_lm.server --model mlx-community/DeepSeek-R1-Distill-Qwen-32B-MLX-4Bit
By default this will launch on port 8080
Generate a global hook directory for git:
mkdir -p ~/.git-hooks
Set it as the global hook location:
git config --global core.hooksPath ~/.git-hooks
Now copy the prepare-commit-msg to ~/.git-hooks/prepare-commit-msg
Make your changes. Add all your files like:
git add .
Then simply run:
git commit
And each file will be sent to the local server and build an overall commit message.