My LoRA Fine-Tuned AI-generated Detector

• Label_0: Represents human-written content.
• Label_1: Represents AI-generated content.

Model Details

• Base Model: intfloat/e5-small
• Fine-Tuning Technique: LoRA (Low-Rank Adaptation)
• Task: Sequence Classification
• Use Cases: Text classification for AI-generated detection.
• Hyperparameters:
  • Learning rate: 5e-5
  • Epochs: 3
  • LoRA rank: 8
  • LoRA alpha: 16

Training Details

• Dataset:
  • 10,000 twitters and 10,000 rewritten twitters with GPT-4o-mini.
  • 80,000 human-written text from RAID-train.
  • 128,000 AI-generated text from RAID-train.
• Hardware: Fine-tuned on a single NVIDIA A100 GPU.
• Training Time: Approximately 2 hours.
• Evaluation Metrics: | Metric | (Raw) E5-small | Fine-tuned | |--------|---------------:|-----------:| |Accuracy| 65.2% | 89.0% | |F1 Score| 0.653 | 0.887 | | AUC | 0.697 | 0.976 |

Collaborators

• Menglin Zhou
• Jiaping Liu
• Xiaotian Zhan

Citation

@inproceedings{dugan-etal-2024-raid,
    title = "{RAID}: A Shared Benchmark for Robust Evaluation of Machine-Generated Text Detectors",
    author = "Dugan, Liam  and
      Hwang, Alyssa  and
      Trhl{\'\i}k, Filip  and
      Zhu, Andrew  and
      Ludan, Josh Magnus  and
      Xu, Hainiu  and
      Ippolito, Daphne  and
      Callison-Burch, Chris",
    booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = aug,
    year = "2024",
    address = "Bangkok, Thailand",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.acl-long.674",
    pages = "12463--12492",
}