64 lines
1.8 KiB
Markdown
64 lines
1.8 KiB
Markdown
---
|
|
license: apache-2.0
|
|
base_model: Qwen/Qwen2.5-Coder-3B-Instruct
|
|
tags:
|
|
- merge-conflict-resolution
|
|
- code
|
|
- qwen
|
|
- qwen2.5
|
|
- coding-assistant
|
|
- git
|
|
- version-control
|
|
- developer-tools
|
|
- code-generation
|
|
- conflict-resolution
|
|
pipeline_tag: text-generation
|
|
library_name: transformers
|
|
language:
|
|
- en
|
|
---
|
|
|
|
# 🌬️ Breeze-3B: AI-Powered Git Merge Conflict Resolution
|
|
|
|
**Breeze-3B** is a specialized coding model fine-tuned on [Qwen/Qwen2.5-Coder-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-Coder-3B-Instruct) to automatically resolve Git merge conflicts with reasoning and context awareness.
|
|
|
|
## 🚀 Key Features
|
|
|
|
- **Intelligent Resolution**: Analyzes merge conflicts and provides reasoned solutions
|
|
- **Multi-Language Support**: Works across Python, JavaScript, Java, C++, and more
|
|
- **Preserves Code Quality**: Maintains general coding capabilities while specializing in conflict resolution
|
|
- **Multiple Deployment Options**: Cloud inference, local GGUF, and Ollama support
|
|
- **Lightweight**: Only 3B parameters - runs efficiently on consumer hardware
|
|
|
|
## 📊 Model Details
|
|
|
|
| Property | Value |
|
|
|----------|-------|
|
|
| **Base Model** | Qwen/Qwen2.5-Coder-3B-Instruct |
|
|
| **Training Data** | 7,165 curated merge conflicts from ConGra dataset |
|
|
| **Fine-tuning Method** | LoRA (rank-8 adapters) |
|
|
| **Parameters** | 3B |
|
|
| **Quantization** | Q4_K_M GGUF available |
|
|
| **License** | Apache 2.0 |
|
|
|
|
### Local Inference
|
|
```python
|
|
from llama_cpp import Llama
|
|
|
|
llm = Llama(model_path="breeze-3b.Q4_K_M.gguf")
|
|
response = llm(f"Resolve this merge conflict:\n\n{conflict}")
|
|
print(response["choices"][0]["text"])
|
|
```
|
|
### Ollama Inference
|
|
``` bash
|
|
ollama run hf.co/SoarAILabs/breeze-3b
|
|
````
|
|
|
|
## Features
|
|
|
|
- Resolves merge conflicts with reasoning
|
|
- Supports multiple programming languages
|
|
- No catastrophic forgetting of general coding skills
|
|
- Works with both cloud and local inference
|
|
|