Reborn to be Wilde: Une approche de transfert stylistique textuel par l’affinage d’un modèle massif de langage
This repository contains the code for the paper Reborn to be Wilde
This project pioneers the finetuning of the mistralai/Mistral-7B-Instruct-v0.1 model for stylistic transfer, aiming to replicate the distinctive writing style of Oscar Wilde with enhanced narrative capabilities. By implementing an authorship attribution classifier based on BertAA, we establish a baseline to measure our success in mimicking Wilde's style. Our ultimate goal is to deceive the classifier into recognizing language model-generated texts as authentic works of Wilde.
No specific installation steps are required to explore this project. However, for running the model and classifier, ensure you have an appropriate Python environment set up with necessary libraries.
Detailed usage instructions will be provided soon. Stay tuned!
While we're not actively seeking contributions, we welcome any interest in the technique we're presenting. Should you wish to experiment with or extend our work, feel free to do so.
This project is licensed under the MIT License - see the LICENSE file for details. The MIT License is a permissive license that only requires attribution back to us and allows for both personal and commercial use.