🌀 Python-to-C++ Code Converter using LLMs

This project offers a user-friendly interface for converting Python code into optimized C++ code using multiple LLMs, including OpenAI, Gemini, and the Hugging Face model NTQAI/Nxcode-CQ-7B-orpo. The interface is built with Gradio, providing a simple and interactive UI. It supports efficient inference using CUDA and 4-bit quantization via the bitsandbytes library for faster performance on compatible GPUs.

🧠 Overview

The notebook does the following:

• Loads a pretrained large language model (NTQAI/Nxcode-CQ-7B-orpo) using Hugging Face Transformers.
• Converts Python code into C++ through prompts and LLM inference.
• Extracts and displays the generated C++ code.
• Hosts a Gradio UI to allow users to input Python code and receive C++ output.

📦 Dependencies

Install the required libraries with:

pip install openai google huggingface_hub google-genai gradio transformers torch bitsandbytes requests accelerate sentencepiece

Ensure you have a CUDA-compatible GPU for best performance.

🔐 Required Tokens

Set your API keys in Google Colab using:

from google.colab import userdata

openai_api_key = userdata.get('OPENAI_API_KEY')
google_key = userdata.get('GOOGLE_API_KEY')
hf_token = userdata.get('HF_TOKEN')

🚀 How It Works

1. A Python sample function (estimating π) is prepared.
2. Prompts are constructed to request the LLM to translate Python to optimized C++ for an M1 Mac.
3. The generate_code_NxCode function sends this prompt to the LLM.
4. The LLM's output is decoded, and C++ code is extracted from markdown-style code blocks.
5. A Gradio interface lets users submit their Python code and receive the C++ version.

🧪 Sample Code Example

Python Input:

def calculate(iterations, param1, param2):
    ...

Expected C++ Output:

#include <iostream>
...

🌐 Web UI

After running the notebook, it launches a Gradio web interface where you can:

• Paste your Python code
• Click "Convert code"
• View the generated C++ in real-time

🛠️ Tech Stack

• Python
• Hugging Face Transformers
• BitsAndBytes for 4-bit quantization
• Gradio (UI)
• Google Colab (for secure key storage)

📌 Notes

• Only supports basic Python-to-C++ logic for now.
• Intended for high-performance translation for M1 Mac systems.
• Output C++ is optimized but may require manual review for production use.