| --- |
| base_model: Qwen/Qwen2.5-Coder-1.5B-Instruct |
| library_name: peft |
| tags: |
| - lora |
| - commit-message-generation |
| - conventional-commits |
| - code |
| --- |
| |
| # Qwen2.5-Coder-1.5B-Instruct — Commit Message LoRA Adapter |
|
|
| LoRA adapter fine-tuned on top of [`Qwen/Qwen2.5-Coder-1.5B-Instruct`](https://huggingface.co/Qwen/Qwen2.5-Coder-1.5B-Instruct) |
| to generate [Conventional Commits](https://www.conventionalcommits.org/) messages from a `git diff`. |
|
|
| Part of a project to build a local, offline `prepare-commit-msg` git hook. |
| Full write-up: https://eliotbas.com/projects/commits-fine-tuning/ |
| Training dataset: https://huggingface.co/datasets/Elib27/commits |
|
|
|
|
| ## Training details |
|
|
| - Method: LoRA (r=16, alpha=32) |
| - Framework: Unsloth + TRL SFTTrainer |
| - Hardware: Google Colab A100 |
| - Dataset: ~11k (diff, commit message) pairs, see [eliotbas/commits](https://huggingface.co/datasets/Elib27/commits) |
|
|
| See the article for full evaluation results (ROUGE-L, structural checks, LLM-as-judge) across all model sizes. |
|
|
