Generates conventional commit messages from git diffs using a fine-tuned Qwen2.5-Coder-3B model. This is a standalone merged model (LoRA adapters merged into base model) ready for direct use.

Model Details

Model: Qwen2.5-Coder-3B fine-tuned on conventional commits
Fine-tuning Method: QLoRA (4-bit quantized, rank=8, alpha=16)
Training Data: 210 real conventional commits from open-source repositories
Model Size: ~3B parameters
Format: Merged model (no base model required)

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer

# Load merged model directly
model = AutoModelForCausalLM.from_pretrained(
    "Pavloffm/qwen-commit-merged",
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("Pavloffm/qwen-commit-merged")

# Generate commit message
diff = """diff --git a/src/main.py b/src/main.py
index 1234567..abcdefg 100644
--- a/src/main.py
+++ b/src/main.py
@@ -1,3 +1,5 @@
+def new_feature():
+    pass
"""

messages = [{"role": "user", "content": f"Generate a conventional commit message for this diff:\n{diff}"}]
inputs = tokenizer.apply_chat_template(messages, return_tensors="pt").to(model.device)
outputs = model.generate(inputs, max_new_tokens=100)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Training

Base Model: Qwen/Qwen2.5-Coder-3B-Instruct
Epochs: 2
Learning rate: 1.5e-4
LoRA rank: 8, alpha: 16
Training examples: 210

License

MIT License