LequeuISIR/AU-clarification_gemma-2-9b-it
The LequeuISIR/AU-clarification_gemma-2-9b-it model is a 9 billion parameter Gemma-2-based instruction-tuned language model developed by LequeuISIR. It is specifically fine-tuned on the GDN-CC dataset for Argumentative Unit Clarification, making it highly effective at rephrasing text segments for clarity, correcting errors, and adding necessary context. This model is designed to clarify opinion text segments in French, ensuring they are comprehensible without external context.
Loading preview...
Model Overview
The LequeuISIR/AU-clarification_gemma-2-9b-it is a 9 billion parameter Gemma-2 instruction-tuned model, specifically fine-tuned for Argumentative Unit Clarification. This model was trained using the GDN-CC dataset and is noted as the primary model used for annotating the GDN-CC-large corpus.
Key Capabilities
- Argumentative Unit Clarification: Excels at rephrasing and clarifying segments of opinion text.
- Error Correction: Automatically corrects spelling, grammar, and syntax errors within the text segments.
- Contextualization: Integrates necessary context from the original text to ensure the clarified segment is standalone and fully understandable.
- French Language Support: Optimized for processing and clarifying French text.
Recommended Use
This model is best utilized for tasks requiring the clarification of argumentative units, particularly in French. It is recommended to deploy it with the vLLM framework for efficient inference. The model's primary function is to take an opinion text segment and its original context, then output a clear, corrected, and self-contained version of that segment. It is particularly useful for enhancing the readability and comprehensibility of citizen consultations or similar textual data.
Citation
For more details on the dataset and its application, refer to the associated research:
@article{lequeu2026gdn,
title={The GDN-CC Dataset: Automatic Corpus Clarification for AI-enhanced Democratic Citizen Consultations},
author={Lequeu, Pierre-Antoine and Labat, L\'{e}o and Cave, Laur\'{e}ne and Lejeune, Ga\"{e}l and Yvon, Fran\c{c}ois and Piwowarski, Benjamin},
journal={arXiv preprint arXiv:2601.14944},
year={2026}
}