Qwen3-Code-Reasoning-4B/README.md

---
license: apache-2.0
datasets:
- GetSoloTech/Code-Reasoning
base_model:
- Qwen/Qwen3-4B-Thinking-2507
pipeline_tag: text-generation
library_name: transformers
tags:
- code-generation
- competitive-programming
- code-reasoning
- programming
- algorithms
- problem-solving
- python
---

# GetSoloTech/Qwen3-Code-Reasoning-4B

A finetuned version of Qwen3-4B-Thinking-2507 specifically optimized for competitive programming and code reasoning tasks. This model has been trained on the high-quality [Code-Reasoning](https://huggingface.co/datasets/GetSoloTech/Code-Reasoning) dataset to enhance its capabilities in solving complex programming problems with detailed reasoning.

## 🎯 Model Overview

This model is a **LoRA-finetuned** version of [Qwen3-4B-Thinking-2507](https://huggingface.co/Qwen/Qwen3-4B-Thinking-2507) with the following specifications:

- **Base Model**: Qwen3-4B-Thinking-2507 (4.0B parameters)
- **Training Method**: LoRA (Low-Rank Adaptation)
- **Training Dataset**: GetSoloTech/Code-Reasoning
- **Training Framework**: Unsloth with QLoRA
- **Context Length**: 4096 tokens (configurable up to 262,144)
- **Model Type**: Causal Language Model with Thinking Capabilities

## 🚀 Key Features

- **Enhanced Code Reasoning**: Specifically trained on competitive programming problems
- **Thinking Capabilities**: Inherits the advanced reasoning capabilities from the base model
- **High-Quality Solutions**: Trained on solutions with ≥50% test case pass rates
- **Structured Output**: Optimized for generating well-reasoned programming solutions
- **Efficient Training**: Uses LoRA adapters for efficient parameter updates


### Dataset Statistics
- **Split**: Python
- **Source**: High-quality competitive programming problems from TACO, APPS, CodeContests, and Codeforces
- **Quality Filter**: Only correctly solved problems with ≥50% test case pass rates

## 🔧 Usage

### Basic Inference

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "GetSoloTech/Qwen3-Code-Reasoning-4B"

# Load the tokenizer and model
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)

# Prepare input for competitive programming problem
messages = [
    {"role": "system", "content": "You are an expert competitive programmer. Read the problem and produce a correct, efficient solution. Include reasoning if helpful."},
    {"role": "user", "content": "Your programming problem here..."}
]

text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
)

model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

# Generate solution
generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=4096,
    temperature=0.7,
    top_p=0.8,
    top_k=20
)

output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist()
content = tokenizer.decode(output_ids, skip_special_tokens=True).strip("\n")
print(content)
```

## 📈 Performance Expectations

This finetuned model is expected to show improved performance on:

- **Competitive Programming Problems**: Better understanding of problem constraints and requirements
- **Code Generation**: More accurate and efficient solutions
- **Reasoning Quality**: Enhanced step-by-step reasoning for complex problems
- **Solution Completeness**: More comprehensive solutions with proper edge case handling

## 🎛️ Recommended Settings

### For Code Generation
- **Temperature**: 0.7
- **Top-p**: 0.8
- **Top-k**: 20
- **Max New Tokens**: 4096 (adjust based on problem complexity)

### For Reasoning Tasks
- **Temperature**: 0.6
- **Top-p**: 0.95
- **Top-k**: 20
- **Max New Tokens**: 81920 (for complex reasoning)

## 🔗 Related Resources

- **Base Model**: [Qwen3-4B-Thinking-2507](https://huggingface.co/Qwen/Qwen3-4B-Thinking-2507)
- **Training Dataset**: [Code-Reasoning](https://huggingface.co/datasets/GetSoloTech/Code-Reasoning)
- **Training Framework**: [Unsloth](https://github.com/unslothai/unsloth)
- **Original Dataset**: [OpenCodeReasoning-2](https://huggingface.co/datasets/nvidia/OpenCodeReasoning-2)

## 🤝 Contributing

This model was created using the Unsloth framework and the Code-Reasoning dataset. For questions about:
- The base model: [Qwen3 GitHub](https://github.com/QwenLM/Qwen3)
- The training dataset: [Code-Reasoning Repository](https://huggingface.co/datasets/GetSoloTech/Code-Reasoning)
- The training framework: [Unsloth Documentation](https://docs.unsloth.ai/)

## 📄 License

This model follows the same license as the base model (Apache 2.0). Please refer to the [base model license](https://huggingface.co/Qwen/Qwen3-4B-Thinking-2507/blob/main/LICENSE) for details.

## 🙏 Acknowledgments

- **Qwen Team** for the excellent base model
- **Unsloth Team** for the efficient training framework
- **NVIDIA Research** for the original OpenCodeReasoning-2 dataset

## 📞 Contact

For questions about this finetuned model, please open an issue in the repository.

---

**Note**: This model is specifically optimized for competitive programming and code reasoning tasks.