DominikMacko/mdok-multiclass

mdok-multiclass detector of machine-generated texts for PAN2025 (available for non-commercial research purpose only - by requesting the access to this model you are agreeing to this condition). It represents the submitted detection model (the best system in the shared task) in Subtask 2: Human-AI Collaborative Text Classification (6 classes).

More info, as well as the training code is available in repo.

Usage

The model is QLoRA fine-tuned Qwen3-4B-Base; therefore, use the latest transformers library supporting it. For the inference, adjust the code.

Cite

If you use the model, code, or any information from this repository, please cite the paper(s):

@misc{macko2025mdokkinitrobustlyfinetuned,
      title={mdok of {KInIT}: Robustly Fine-tuned {LLM} for Binary and Multiclass {AI}-Generated Text Detection}, 
      author={Dominik Macko},
      year={2025},
      eprint={2506.01702},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2506.01702}, 
}

@misc{macko2025increasingrobustnessfinetunedmultilingual,
      title={Increasing the Robustness of the Fine-tuned Multilingual Machine-Generated Text Detectors}, 
      author={Dominik Macko and Robert Moro and Ivan Srba},
      year={2025},
      eprint={2503.15128},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2503.15128}, 
}

Funded by the EU NextGenerationEU through the Recovery and Resilience Plan for Slovakia under the project No. 09I01-03-V04-00059.