| # SeqXGPT - Sentence-Level AI-Generated Text Detector | |
| A trained SeqXGPT model for detecting AI-generated text at the sentence level. | |
| ## Model Description | |
| This model implements the SeqXGPT architecture from the paper ["SeqXGPT: Sentence-Level AI-Generated Text Detection"](https://arxiv.org/abs/2310.08903). | |
| ## Training Results | |
| | Metric | Score | | |
| |--------|-------| | |
| | Sentence Accuracy | 95.6% | | |
| | Macro F1 Score | 95.6% | | |
| | Word Accuracy | 95.5% | | |
| ## Supported Labels | |
| - `gpt2`: GPT-2 generated | |
| - `gptneo`: GPT-Neo generated | |
| - `gptj`: GPT-J generated | |
| - `llama`: LLaMA generated | |
| - `gpt3re`: GPT-3 rewritten | |
| - `human`: Human written | |
| ## Usage | |
| ```python | |
| import torch | |
| from model import ModelWiseTransformerClassifier | |
| # Load model | |
| model = torch.load("seqxgpt_transformer.pt") | |
| model.eval() | |
| # Inference requires feature extraction from 4 LLMs (GPT-2, GPT-Neo, GPT-J, LLaMA) | |
| # See the original SeqXGPT repo for full pipeline | |
| ``` | |
| ## Citation | |
| ```bibtex | |
| @misc{wang2023seqxgpt, | |
| title={SeqXGPT: Sentence-Level AI-Generated Text Detection}, | |
| author={Pengyu Wang and Linyang Li and Ke Ren and Botian Jiang and Dong Zhang and Xipeng Qiu}, | |
| year={2023}, | |
| eprint={2310.08903}, | |
| archivePrefix={arXiv}, | |
| primaryClass={cs.CL} | |
| } | |
| ``` | |