Training a Local LLM (Qwen3.5-2B) to Generate Git Commit Messages Using MLX + LoRA

Sun, 08 Mar 2026 00:00:00 +0000

I trained a small local LLM to generate conventional Git commit messages directly from diffs.The entire setup runs locally on a MacBook using MLX + LoRA, and training took about 25 minutes.Here is how the pipeline works.

TL;DR

Fine-tuned Qwen3.5-2B-Base locally using MLX + LoRA
Built a dataset of ~1400 commit diffs
Training took ~25 minutes on a MacBook M4 Pro
Final model generates conventional commit messages from diffs
Git Repo: https://github.com/nithinputhenveettil/mlx-llm-commit-message-finetuning

Example:

What made you choose your industry? ✏️

Sat, 01 Feb 2025 00:00:00 +0000

My Journey into Software Engineering!!

“As far back as I can remember, I always wanted to be a software engineer.”

Just kidding, guys 😅. Nope, this is not the case for me. I am just quoting the famous dialogue from Goodfellas.

For me, I was introduced to programming in the year 2009 when I joined my higher secondary schooling. It was C++, running in its famous blue IDE with large fonts and everything, that too on an old Windows XP machine. At that time, I didn’t have a PC at home. To be honest, I didn’t find programming that amusing in those days. I was trying to learn code by heart instead of understanding the language itself.

Nithin Puthenveettil

Training a Local LLM (Qwen3.5-2B) to Generate Git Commit Messages Using MLX + LoRA

TL;DR

What made you choose your industry? ✏️

My Journey into Software Engineering!!