SBD04 - Custom Coding AI Assistant

A fine-tuned language model optimized for programming tasks, built on Qwen2.5-Coder-1.5B-Instruct architecture.

📦 Models

File	Size	Description
sbd04-qwen2.5-coder-1.5b-q4_k_m.gguf	~1.1 GB	Recommended - Quantized Q4_K_M, fast inference
sbd04-qwen2.5-coder-1.5b-f16.gguf	~3 GB	Full precision F16

🚀 Quick Start

LM Studio

1. Download sbd04-qwen2.5-coder-1.5b-q4_k_m.gguf
2. Open LM Studio → My Models → Add Model → From File
3. Select the GGUF file
4. Set context length to 4096+ and start coding

llama.cpp

./main -m sbd04-qwen2.5-coder-1.5b-q4_k_m.gguf -p "Write a Python function to parse JSON" -n 512

Python (llama-cpp-python)

from llama_cpp import Llama

llm = Llama(model_path="sbd04-qwen2.5-coder-1.5b-q4_k_m.gguf", n_ctx=4096)
output = llm("Write a binary search in Python", max_tokens=512)
print(output['choices'][0]['text'])

🎓 Training Details

• Base Model: Qwen2.5-Coder-1.5B-Instruct
• Method: LoRA fine-tuning (r=16, α=32)
• Dataset: 60 coding Q&A pairs covering Python, JavaScript, TypeScript, SQL, Git, Docker, REST APIs, Design Patterns, Debugging, Performance, Security, and more
• Epochs: 3
• Trainable params: ~18.5M (1.18% of model)
• Max context: 1024 tokens

💡 Example Prompts

"Write a Python decorator for timing functions"
"How do I implement a binary search tree in Python?"
"Review this code for bugs: [paste code]"
"Explain async/await in JavaScript"
"How to prevent SQL injection in Python?"
"Write a Dockerfile for a FastAPI app"
"What's the difference between INNER JOIN and LEFT JOIN?"
"How to use TypeScript generics?"
"Optimize this slow Python loop: [paste code]"
"Write a pytest fixture for database testing"

📄 License

Apache-2.0

🔗 Links

• Hugging Face: SBDO1/SBD04
• Created by: SBDO

---

Built with ❤️ for developers everywhere