AI Detection in Code

Description

This project aims to detect the usage of AI in coding responses by predicting an AI-detected score. The model compares candidate answers with responses generated by various AI models (GPT-4, GPT-4 Turbo, and GPT-3.5 Turbo) to determine the likelihood that a response was AI-generated.

Requirements

To set up and run this project, follow these steps:

1. Python: Ensure you have Python 3.9.10 installed (or a compatible version).

2. Create a Virtual Environment (recommended):

   python -m venv venv
   source venv/bin/activate  # On Windows use `venv\Scripts\activate`
   

3. Install Dependencies:

   pip install -r requirements.txt
   

Notebooks

The following Jupyter Notebooks are included for reference and are optional to run since the training has been completed and models have been pushed to the hub:

• data_preparation.ipynb: Prepares the dataset for model training.
• finetune_LLM.ipynb: Fine-tunes the language model.
• finetune_similarity.ipynb: Fine-tunes the similarity model.
• finetune_regression_model.ipynb: Fine-tunes the regression model.

Testing the Model

To test the fine-tuned model, use the provided notebook:

• deployment.ipynb: Tests the fine-tuned models to validate the performance.

Files

• data.csv: The dataset used for training the models.
• vector_db.pt: A mini vector database for similarity model inference.

Documentation

For a detailed explanation of the process, methodologies, and choices made, please refer to the documentation.md file.

Contact

• Email: ikram.djeghali@gmail.com
• HuggingFace : https://huggingface.co/wasabibish